Culture and Development Jon R. Jellema* Department of Economics, U.C. Berkeley December 2008 Job Market Paper

Abstract Whether culture affects development is one of the most fundamental questions in economics, but sample, measurement, and direction-of-causation issues hinder empirical analysis. Making use of advances in empirical anthropology and population genetics, I provide robust solutions to these problems. I assemble measures of cultural behavior collected systematically from more than 1200 anthropological case studies. I describe the generation of cultural variety without invoking previously existing institutions or tendencies. I exploit the parallel random mutation and long-term persistence of genetic and cultural information in an instrumental variables framework where I demonstrate that predictable variation in neutral genetic information (and not genetic inheritance of social information) provides a valid and powerful instrument for culture. I show that within this instrumental variables framework class stratification, inheritance rights, and other cultural technologies can explain up to 115 percent of a standard deviation in output, or approximately the size of the gap in per-capita GDP between Thailand and Ireland.

*

Contact information: 508-1 Evans Hall #3880; University of California, Berkeley; Berkeley, CA 94720-3880. Email: [email protected].

1.

Introduction

Economic exchange is a social transaction. The outcomes resulting from it, expectations surrounding it, and its prescribed structure cannot be isolated from the shared norms and mandated behaviors that accompany all social transactions. The role of culture in economic development, in other words, is fundamental. However, social behaviors, rules, norms, and standards can logically be determinants or results of economic exchange, making empirical identification of the developmental effects of culture problematic. Culture itself is difficult to measure as it requires observation of social interaction which is multi-faceted and not amenable to summary by indicators like price or quantity; neither are data capturing singular opinions or beliefs precise guides to social activity. In this paper, I assemble two global datasets that offer novel and robust solutions to these identification and measurement problems in order to demonstrate empirically the causal relationship between cultural behavior and economic development. Specifically, I show that cultural practices promoting division of labor and specialization, informal education and research and development, or property rights lead to cross-sectional increases of up to 70 percent of a standard deviation of economic development. Increases in all of them simultaneously lead to increases of up to 115 percent of a standard deviation of economic development. Cross-sectional differences in current real GDP per capita roughly this size can be found between Thailand and Ireland, the Czech Republic and Belgium, Colombia and Japan, or Bhutan and Chile, for example. These relationships remain visible within geographic regions and production technologies and each cultural practice remains individually predictive when others are held constant. To measure culture convincingly, I use observations from detailed ethnographies covering more than 1200 populations worldwide. This dataset is richer in both detail and scope, more objective, less prone to perception bias, and contains less non-random noise than other cultural datasets. Culture is observed closely and recorded as behavior rather than belief. The sample of populations is world-wide and includes groups and areas, like Pacific Island societies, often missing from empirical studies; there is no oversampling of populations by income, geographic location, or colonial history. I describe the discovery and long-term persistence of cultural practices among populations choosing new environments to in which to settle. Variation in cultural practices is generated by problems like the management of group resources and the labor of individual 1

members. The fixed environment, choice of subsistence activity, and risk stemming from the interaction of these two variables force social decisions regarding how endowments will be managed and what goals they should serve. Random mutation in social behavior is equally important, so similar endowments do not necessarily generate similar cultures.1 All of these variables (endowments, risk, production) in addition to culture determine economic development simultaneously. The event generating variation in these constraints under which social practices develop is the serial migration of modern humans across the globe that began roughly 100,000 years before present and ended roughly 10,000 years before present. This event also left a remarkable signature on the human genome visible at the population level: genetic variety, or the population frequency of potential genomic possibilities, is inversely proportional to distance from East Africa, the original start line for this era of migration that eventually brought humans to the extreme south of South America and virtually every habitable locale in between. This information has been measured across a large portion of the human genome that does not code for the production proteins and is not associated with observable behaviors or physical characteristics. It records only neutral genetic diversity, or that portion of overall genetic diversity that is not under natural. Because it is not a cause of any observable behavior or characteristic, it is not an object of choice or optimization for economic actors. Variation in this information is analogous to results from successive random draws (at the population level) without replacement from the original pool of genetic material, leaving populations near the source with the most variety and those farthest away with the least. I demonstrate a significant and robust correlation between this candidate instrument and specific cultural technologies. Since both genetic and cultural information are transmitted with modification from parent to offspring and persist across generations, this result is expected: populations with similar ancestors are more alike in all vertically-transmitted information. I exploit the correlation and the assumption that variation in genetic heterogeneity is exogenous to income generation to pursue an instrumental variable strategy that demonstrates empirically the main hypothesis of this paper: culture creates economic incentives and human capital that affect development in a robust and economically significant manner.

1

Though developed consciously and purposefully, cultural outcomes are not always predictable, nor is the decision to adopt any norm always observable. In both senses, the discovery of cultural practices is partially random.

2

More generally, I provide empirical support for the hypothesis that unpredictable cultural norms create long-lived preferences and incentives.2 I affirm that luck in these historical endowments can alter the course of economic development.3 From a policy perspective, the key finding is that informal, uncodified, and often invisible institutions create economic incentives. This suggests that formal institutions should be adapted to local conditions rather than transplanted wholesale. The plan of the paper is as follows: Section 2 reviews the literature on culture and economic development, focusing on empirical treatments with careful identification strategies. Section 3 provides historical background and description of the mechanism linking migration, cultural behaviors, genetic variety, and development. Section 4 describes data sources. Section 5 presents empirical specifications and results, brief discussions of the pathways between cultural behaviors and development, and examples of the cultural behaviors at work. Section 5 also includes tests of the validity of the instrument and robustness testing. Section 6 offers concluding remarks. Extended robustness testing, detailed discussion of the pathways from cultural technologies to development, and detailed examples of cultural behavior at work are found in the appendices at the end. 2.

Literature Review

Empirical analysis of the link between cultural variables and economic outcomes is a relatively new research program. Some early examples like Knack and Keefer (1997), Temple and Johnson (1998), and Hall and Jones (1999), while they do not explicitly observe culture, use indices of ―social infrastructure‖ which likely contain latent elements of culture. These early empirical analyses are in agreement: variation in social infrastructure predicts variation in economic outcomes. Roland and Jellema (2006) produce evidence that culture and political institutions are complements in income generation and that culture remains a significant predictor of income when other institutions are held constant. Using historical eras or chronologically distant events as a source of plausibly exogenous variation in formal or informal institutions is also a relatively new avenue of research. Acemoglu, Johnson, and Robinson (2001, 2002) exploit variation in the disease environment confronting colonial settlers arriving in the 1500s to identify variation in current levels of the risk 2 3

See Akerlof and Kranton (2005), Logan and Rhode (2008), or Roland (2004). See Diamond (1997) for physical technologies and Tabellini (2007) for cultural technologies.

3

of expropriation. Nunn (2008) finds that variation in the intensity with which nations participated in the 13th through 15th century slave trades predicts variation in current income.4 Galor and Moav (2007) suggest that years since the introduction of agriculture can explain contemporary variation in life expectancy; they hypothesize that the Neolithic revolution brought cultural and biological changes which in turn made longer life expectancy not only possible but optimal.5 Recent examples combining both explicitly cultural variables and plausibly exogenous sources of historical variation include Tabellini (2007), Licht, Goldschmidt, and Schwartz (2007), and Guiso, Sapienza, and Zingales (2008). The first two combine cross-national observations of beliefs or opinions and an identification strategy that uses variation in linguistic rules determined by ―distant traditions‖.6 Each describes an interaction between a set of beliefs (as recorded by survey) and formal institutional outcomes like the rule of law or the quality of government. The subsequent interaction between institutions and economic outcomes is not statistically tested but is assumed given the wealth of evidence available from other studies. The latter finds that historical experience with independent city states in Italy is associated with greater social capital (measured by civic participation). The authors hypothesize that the cultural norm of cooperation reaffirmed by that historical experience has persisted and produces higher social capital and income today. Versions of the genetic data underlying this paper‘s proposed instrument have been used before. Two studies, Spolaore and Wacziarg (2009) and Ashraf and Galor (2008), estimate a direct effect of genetic dissimilarity on economic outcomes.7 With a condensed version of the genetic data, Spolaore and Wacziarg (2009) take pairwise genetic differences at the country level and estimate the contribution of these differences to pairwise differences in income. They 4

Though institutions are not statistically identified or tested in Nunn (2008), the author suggests them as intermediaries through which slave-trade participation might operate on current income. 5 Olson and Hibbs (2005) also begins at the Neolithic revolution but does not include an independent effect of institutions. But see also Putterman (2008), which replicates Olson and Hibbs (2005) with corrections for the diffusion of populations and technologies, including social practices. 6 See also Alesina and Fuchs-Schündeln (2005), Giuliano (2007), Fernandez and Fogli (2007), or Munshi and Wilson (2008), all of which use the not-too-distant traditions of first-generation immigrants as instruments for the cultural practices of second- or third- generation offspring of those immigrants. These later generations receive a cultural package developed for a substantially different environment and variation in outcomes among them is correlated with this cultural variation. 7 One of the first examples of genetic information used as an instrument in a two-stage least squares framework is Fletcher and Lehrer (2008). The authors use variation in the presence of specific genes known to impact multiple health outcomes to isolate the effects of poor health on educational outcomes. Though I use genetic information that does not lead to differences in phenotype (observable characteristics, attributes, or behaviors), my empirical strategy is much the same as Fletcher and Lehrer (2008) as I rely on the inability of economic actors to make genetic information a choice variable.

4

hypothesize that pairwise genetic dissimilarity is a proxy for differences in ―characteristics, including cultural traits‖ which act as barriers to the diffusion of innovation. Ashraf and Galor (2008) estimate the direct effect of expected variation in genetic material on national population densities in 1500 or earlier. The authors take genetic diversity as a proxy for intra-population cultural dissimilarity and hypothesize two competing effects on population density: genetic diversity hinders the transmission of ―society-specific human capital‖ but encourages the ―accumulation of universally-applicable human capital‖.8 This leads to a non-monotonic and indeterminate effect of genetic diversity on population density. This paper, with an emphasis on the local generation of observable behaviors and their effects on development, offers the following improvements to analyses discussed above: the candidate instrument is unobservable by economic actors; I do not rely on previously existing cultures, institutions, or external innovations to explain developmental or cultural variety; and I am able to show which cultural behaviors matter for output. 3.

Instrument, Hypothesis and Historical Background

Economists have long suspected that culture and development covary - see Smith (1759), Weber (1905), or Bowles (1998) for a modern summary - and empirical analyses have lent broad support to this hypothesis, but measurement and identification issues are inescapable. In Sections 5 and 6 I provide a discussion of why certain cultural technologies might cause development (see also Appendices B and C), describe cultural measurement, and provide additional details on the candidate instrument, including its ability to identify exogenous cultural variation. Following directly below is a description of the candidate instrument and the mechanism generating both cultural and instrumental variation. 3.1.

Candidate Instrument

The instrument is population-level variety in neutral genetic information.9 Along the human genome there are many thousands of sites where different versions of a gene, nucleotide group,

8

Spolaore and Wacziarg (2009) and Ashraf and Galor (2008) both suggest a relationship between variation in genotype, or the unobservable genetic makeup of a person, and variation in the observable behaviors they believe lead to economic growth, but these relationships are not tested empirically. 9 Effectively, this means I use variation-in-variation to identify variation in culture.

5

or a single nucleotide10 can potentially occur. An individual carries only one variant at any site where multiple variants are possible. If an individual is carrying a particular variant in the section of the human genome that determines blood type, for example, then she will end up with blood type A. A different variant at the same site in another individual leads to blood type B and another possible variant to blood type O. Three individuals, each with a different variant, will carry three different blood types. Across a population all variants are observed more or less often and frequencies can be calculated. Individuals with varieties that produce blood types A and B are frequent in the Ainu of Japan and nearly nonexistent in the Zuni of New Mexico, for example: population frequencies for A, B, and O blood types are approximately 26, 20, and 54 percent and 1, 6, and 93 percent for the Ainu and Zuni respectively (Cavalli-Sforza, Menozzi, and Piazza 1994). At this site, therefore, the Ainu are more heterogeneous than the Zuni: two randomly selected Ainu individuals will more often carry different blood types. Genetic information that is not advantaged selectively exhibits varying population frequencies due to what is essentially sampling error: populations carry only those varieties their ancestors carried.11 More precisely, they carry only those varieties transmitted by reproductive individuals within the ancestral group. Endogamous populations with different ancestors carry different sets of gene variants and not every ancestor has contributed genetic information. This compound sampling error leads to variation at the population level even in the absence of clearly advantageous varieties. Blood type is not selectively advantageous, but sampling error and endogamy have created divergent gene frequencies. I use variety in genetic information that does not confer a selective advantage averaged over up to 1000 multi-variant sites. Each site is not known to specify the production of any protein nor is it associated with any observable behaviors or physical characteristics.12 Imagine each population is associated with 20 pages, drawn randomly, one for each population member,

10

Nucleotides are the building blocks of DNA, which is itself the material of which genes are made. Nucleotides are the discrete chemical compounds that line up along the familiar double helix in which DNA is arranged. 11 Population variation may also arise from natural selection, or the differential reproductive success of genomes with or without certain varieties. Sickling of red blood cells, for example, occurs more frequently in populations living in environments where resistance to malaria confers a reproductive advantage. Sexual selection may also produce variation if observable characteristics or behaviors increase the likelihood of sexual reproduction by increasing attractiveness to mates. 12 Though a large portion of the human genome is ―non-coding‖ in this way, these sequences are not necessarily all non-functional.

6

from a phonebook containing all possible 10-digit phone numbers.13 When two members reproduce, each of their pages is copied automatically and given to offspring. These phonebooks have two special properties: first, the numbers, if dialed, would not reach anyone – they are ―non-coding‖. Second, neither the numbers nor the copying are observable by any group member. The analogous variation I might propose as an instrument would not be the presence or absence of any particular number or group of numbers. Instead, the population frequency of varieties would be observed and from that count a summary measure that describes overall heterogeneity in phone numbers (how many different numbers are observed in the fourth digit, for example, and how often?) would be calculated and averaged over all digit locations. Neutral genetic variety from sampling error is distributed around the world in a predictable manner (see Sections 3.5 and 4.2). I calculate this distribution for a sample of populations and use it as the instrument.14 It is unobservable, does not confer a selective advantage, and was not a choice variable. Taking an average over several hundred sites effectively dampens the signal from any one site, so even if some sites are discovered to be under selection, the average information remains neutral.15 Genetic information, like cultural information, is reproduced and persists by sexual reproduction and vertical transmission from parent to offspring. In other words, each time the non-coding phonebooks are passed on, an additional set of phonebooks with cultural codes are also replicated and internalized. With observations from both sets of information for a sample of populations, it is possible to identify the cultural codes that have persisted by vertical transmission across generations from their correlations with persistent neutral genetic codes (see Section 3.5). 3.2.

Historical Diffusion of Human Populations

Figure 1 describes routes taken by modern humans to all currently inhabited areas of the globe. The dates shown are estimated arrival times in years before present; confidence intervals on proposed dates are not narrow. For example, the evolutionary events leading to the establishment of modern humans in Africa are thought to have occurred between 200,000 and 13

Following the rules for US exchanges (no zeros or ones in the first digit of the area code or prefix, no quadruple zeros in the suffix), there are just under 6.4 x 109 possible phone numbers. 14 It is important to note that this variety is always based on genetic information (which nucleotide varieties are present) and never on the physical outcomes (phenotypic variety) that genetic information sometimes instructs. 15 This is true also for gene flow, or the exchange of genetic material between neighboring populations through exogamy: a signal that happens to be the product of exogamy will be dampened by averaging over many sites.

7

150,000 years ago; expansion out of Africa between 100,000 and 65,000 years ago; and the first arrival of humans in North America between 40,000 and 10,000 years ago (Mellars 2006). Intracontinental migration dates and arrival times are similarly imprecise (Lahr and Foley 1994). FIGURE 1 – ANCIENT MIGRATIONS of MODERN HUMANS

Note: Figure 1 shows hypothesized routes to and dates of arrival in all currently inhabited areas. Dates shown are years before present and are inexact. Adapted from Cavalli-Sforza and Feldman (2003).

There are two important features of these migrations the empirical analysis exploits. The first is the ―Out of Africa‖ or ―Recent Single Origin‖ hypothesis, which states that there was one site-specific beginning to the diffusion and expansion of modern humans; the available evidence suggests the site was East Africa (Harpending and Rogers 2000). The second concerns the composition of successive migrating groups: each outmigrant group was a genetically non-representative subset of the stay-at-home population (Prugnolle, Manica, and Balloux 2005). A recent single origin links every human population to the founding population while serial 8

selection of genetically non-representative migrating groups produces population variation in the human genome.16 3.3.

Cultural Innovation and Maintenance

Each migrating group confronted a distinct and unfamiliar environment requiring new strategies and technologies for subsistence and reproduction.17 These strategies and techniques include ideas about how members will relate to one another socially and familiarly, collective actions and goals, taboos, leisure, and myriad other norms and conventions all of which can properly be called culture. Some of these may be explicitly optimal solutions to unambiguous problems,18 some may be the result of extensive trial and error, and some may have occurred serendipitously.19 All locally-developed solutions need not have been a result of local inputs and local experience alone. Human capital (including culture) and physical capital (tools, production techniques) acquired during earlier eras may have been locally useful.20 This ability to reoptimize and co-adapt to a broad range of ecological niches is unique (historically if not biologically) to humans, and the available evidence suggests that the elements necessary for this relentless innovation were in place before populations moved from Africa.21 Once developed, local solutions to social problems will be transmitted from parent to child and generation to generation (Bisin and Verdier 2001; Cavalli-Sforza and Feldman 1981). The vertical transmission of local cultural solutions results in a stable set of cultural values which are slow to change (Roland 2004). Some of the cultural technology may prove advantageous as 16

Though evidence from archaeology, linguistics, anthropology, and population genetics supports an East African origin and serial migration, a multi-site model of the origin of modern humans has not been conclusively ruled out. Though it would make interpretation of the patterns present in population genetic information less straightforward, a multi-site model would not invalidate either the hypothesis of cultural innovation or the empirical results discussed below. 17 Lahr and Foley (1998) argue that these ―myriad local histories‖, rather than population exchange, are responsible for most of the linguistic, cultural, morphological, and genetic diversity among modern human populations. 18 Also likely is toleration of ideas that are not welfare-improving but that are maintained by collective action or coordination failures (Olson 1965). 19 Both Roland (2004) and Tabellini (2008) accommodate randomness or luck in the history of ideas, as no norms concerning human interaction are accepted everywhere. Diamond (1997) leaves room for chance in the worldwide distribution of subsistence practices and technologies by noting both that the most productive subsistence strategy, sedentary agriculture, did not arise first in the areas most suited for it and also that the original development of subsistence agriculture was probably accidental. 20 Roland (2004) defines the accumulated stock of embodied knowledge as technology. I will also treat cultural strategies as part of the technological endowment that determines how a given society functions. 21 The cognitive, neurological, cultural, and technological changes that occurred before expansion out of Africa may be directly associated with the evolution of anatomically and genetically modern populations (Mellars 2006).

9

populations grow and begin to produce above subsistence levels.22 Larger populations and production above subsistence in turn make possible specialization, intra-group trade, and groupwide improvements in production scale and scope, all of which lead to greater incomes or levels of development. 3.4.

Schematic: Cultural Innovation, Persistence, and Income

The preceding arguments concerning human diffusion, cultural innovation, cultural maintenance, and economic development are captured in the following set of equations governing the distribution of population income over the long run:

Yj = k(Cj, Ej, Aj, H) + εj,

(1)

Cjt =

(2)

H = m(

, S).

(3)

In equation (1), development Y is determined by cultural strategies C, environment E, subsistence activity A, and technology H, which is an accumulated stock including physical and human capital. Location is indexed by j, generations by t ; ε, η, and υ are all random shocks; σ is the share of output devoted to durable physical capital and α is a constant that describes the rate at which the stock of capital decays. S represents time elapsed since settlement and captures the gradual improvements in technology that come from extensive trial and error or familiarity with the local environment, including knowledge of other groups Equation (2) describes the mechanism generating the innovation and diffusion of cultural strategies. Migration brings a people to a new homeland, where cultural strategies develop randomly under constraints to meet the demands of social organization and subsistence production in new environments. Once migration has ended and location is fixed, transmission with modification across generations leads to relative stability in cultural technologies. The distribution of culture generated by equation (2) includes error terms capturing random shocks to culture that might, for example, be generated by contact with a neighboring group, the discovery of a new food source, a re-interpretation of a sacred myth, or any event that 22

As discussed below, culture is determined at least partially by randomness in the generation and adoption of ideas. The generation of ideas is analogous to the generation of mutant genotypes in the sense that it is unpredictable and unobservable..

10

leads to either the insertion of new or the deletion of old cultural strategies. The presence of these shocks means that two groups with similar initial endowments E, A, H, andY may not end up culturally similar. Shocks ε, η, and υ may contain common elements. Consider a prolonged drought which reduces production and also leads to the adoption of new methods by which authorities are chosen. A common shock produces predictable (reduced output) and unpredictable (a new cultural norm) outcomes. The unpredictable element is perpetuated by the vertical transmission of cultural behaviors where it may be further modified.23 The cultural innovation may eventually affect income if, in addition to specifying a new mode of political organization, it creates incentives to produce, or leads to increased property security. Persistence leads to stability in technology H. Cultural strategies are re-generated each period t so depreciation α of the stock of accumulated knowledge is counterbalanced by generational renewal. More distant technology depreciates faster: α’(t) is assumed to be less than zero. Figure 2 shows an example of these processes: the distribution of evil-eye belief for populations in the empirical analysis (see Section 4.2.1). Evil-eye belief is the belief that a look, touch, or verbal expression of envy or excessive praise can cause material harm like sickness, loss of vitality or even death.24 In Sub-Saharan Africa, it is distributed evenly, though its absence from the west is conspicuous. In the Circum-Mediterranean (North Africa, Europe, and the near Middle East), evil eye belief is ubiquitous. Then, moving through East Asia, the farthest east where it is seen is Bhutan; it is absent from Sri Lanka through Southeast Asia, China, Japan, and far east Russia. In the Pacific, where it occurs with the least frequency, it is seen at least as far east as Fiji. In North America, evil eye belief is confined to the west coastal corridor and in South America it has not spread east or south. Figure 2 shows discontinuities in both the frequency and spatial distribution of behavior which suggest that migration and endogamy generate unpredictable variation in cultural strategies. Significant physical boundaries or long distances can produce clusters of groups with similar technologies, but groups that have crossed such boundaries do not predictably choose those technologies their ancestors had. 23

Durham (1991) provides several examples of the transmission with modification of cultural behaviors adopted after environmental shocks to subsistence production. Grimm and Klasen (2008) find that local migration induces change in formal institutions. 24 Evil-eye belief is not predictive of development – see section 5.5.1.

11

FIGURE 2 – EVIL EYE BELIEF WORLDWIDE

Circum-Mediterranean

Sub-Saharan Africa

East Asia

12

FIGURE 2 CONTINUED– EVIL EYE BELIEF WORLDWIDE

The Pacific

South America

North America

Note: Evil eye belief (dark circles) for 175 societies in Sub-Saharan Africa, Circum-Mediterranean, East Asia, the Pacific, North America, and South America (consecutively from top left). See the accompanying text for definitions.

13

Equations (2) and (3) highlight the tendency of cultural strategies to evolve randomly under local constraints at any j and then to persist over long periods of time after location has become fixed. Equation (1) proposes that these same behaviors affect production. Genetic information is conspicuously absent from the schematic model in equations (1) through (3) by design: there is no direct mechanical link or singular mapping from genetic to cultural information. Nonetheless there is robust global correlation between the two sets arising from evolution under common constraints. 3.5.

Human Diffusion and the Generation of Cultural and Genetic Diversity

Neutral genetic variation is correlated with variation in social behaviors via like processes of innovation, diffusion, and permanence. The particular manner in which early migrations were achieved (serially, by genetically non-representative subsets, from a single origin) led to founder effects and genetic drift, resulting in long-lived variation in overall genetic information at the population level (Lahr and Foley 1998; Ramachandran et al. 2005). A ―founder effect‖ describes the loss of genetic variation that occurs when a new society is established by a small number of individuals who by necessity carry only a subset of genetic information from the originally available pool. ―Genetic drift‖ is the probabilistic deviation of genetic information due to random variations in which members of any population actually reproduce. During the long history of human diffusion, ―Out of Africa‖ and serial migration led to repeated founder effects and drift within endogamous founding populations; the effect is a regular decrease in genetic heterogeneity from populations near East Africa to populations further away (Li et al. 2008). As groups migrated, coalesced, and sent new migrants further along, each successive group carried with it increasingly smaller subsets of genetic information from the originally available pool. Once settled, genetic drift operated on each endogamous population, causing further divergence in genetic information from original populations.25 Populations with less divergence in neutral genetic information have shared common ancestors more recently. When migrants m leave a sending population s, founder effects and genetic drift eliminate some portion γm of existing genetic variants. A subsequent migration of

25

There was likely further adjustment in overall genetic variability depending on the amount of gene flow achieved by sexual reproduction between neighboring populations. However, genetic variability was only partially replenished by gene flow from neighboring groups, so the set of local variants by and large remained local and were perpetuated across generations through local, endogamous sexual reproduction (Lahr and Foley 1998).

14

m′ from s′ (= m) again takes away γm′ of genetic variants26 at s′, leaving m′ and s′ closer genetically than m′ and s. As neutral genetic information evolves among distant but related populations, those more closely related ancestrally will be closer genetically. As with genetic information, isolated populations that more recently shared a cultural information set should be more culturally similar today. Each migratory move may generate cultural mutation because moves are accompanied by new environments and the passage of generational time.27 But cultural behaviors are renewed generationally via vertical transmission from parent to offspring, so on average they should be most alike in populations with more recent common ancestors.28 More distantly related populations with more shocks, generations, or migratory stops between them should be less alike culturally.29 The instrumental variable strategy relies on a correlation between variation in genetic information and variation in cultural behavior: two ancestrally close populations will be more similar culturally and genetically than two ancestrally distant populations. Figures 3a and 3b demonstrate this correlation for a group of populations in Asia and the Pacific. Figure 3a shows the distribution of postpartum sexual taboos in this region and Figure 3b the relationship between pairwise cultural and genetic differences for the populations in 3a. Pairwise genetic distance is based on differences in neutral variety and summarizes the length of time a pair has been isolated from each other: the greater the genetic distance, the greater distance to a common ancestor (Cavalli-Sforza, Menozzi, and Piazza 1994). The greatest differences in cultural behaviors are found in those pairs who have been isolated from each other the longest, and the most similarity in cultural behavior among those pairs with close common ancestors.

Migrants m′ receive genetic information available from s′; some genetic information from s is not available to m′. Cultural information is not always transmitted or maintained with great fidelity. This should caution against any hypothesis that asserts a causal link between genes and cultural behavior. It should also, in an empirical treatment of equation (1) with genetic information as an instrument for culture, bias first-stage coefficients towards zero. 28 This is borne out empirically both in this study (see Figure 3a & 3b and section 5) and indirectly in other studies. Fernandez and Fogli (2007), Alesina and Fuchs-Schündeln (2005), and others document pervasive cultural similarities between migrants and ancestors. 29 This remains true if C evolves wholly randomly. Consider a random walk with each migrating population receiving ancestral culture plus a zero-expected value random shock (Cm = Cs + η, P(η>0) = ½, P(-η) = ½): in expectation, m′ and s′ (=m) will be more culturally similar than m′ and s. Spolaore and Wacziarg (2009) note this relationship in forming expectations regarding the empirical correlation between distances in genetic information and distances in other ―vertically-transmitted characteristics‖. 26 27

15

FIGURES 3a and 3b – GENETIC and CULTURAL DIFFERENCES in EAST ASIA and PACIFIC 3a

3b 0.4

mean genetic distance; bar endpoints at +/− 1 st. dev.

Genetic Distance

0.3

0.2

0.1

0 0

1

2

3

4

5

Absolute Difference in Postpartum taboos Note: Figure 3a shows the distribution of postpartum sexual taboos in East Asia and the Pacific. Postpartum taboos take 6 categorical values, from no taboo to prohibitions longer than 2 years. Larger/darker dots are longer taboos. Figure 3b shows mean genetic distance for those pairs of populations with postpartum taboo differences (in absolute value) along the x-axis.

4.

Data

The data described here permit an instrumental variables estimation of equation (1), which describes the effect of culture on levels of development, for a world cross-section of populations. 4.1.

Cultural Data

Cultural technologies are observed and recorded in the Standard Cross Cultural Sample (SCCS) (Murdock and White 1969). The observations in the SCCS are extracted and coded from the Ethnographic Atlas (Murdock 1967), a compilation of over 1200 ethnographies that collectively cover virtually all modern and many pre-modern societies. The SCCS selects populations from the Atlas, each pinpointed to the smallest identifiable subgroup, to achieve a distribution of human groups with independent histories, homes, and

16

cultures. Murdock and White (1969) describes the sample selection mechanism as follows: the original universe of over 1200 well-described populations was partitioned into ―groups of societies with cultures so similar…that no world sample should include more than one of them.‖ These clusters were then grouped into roughly 200 sampling provinces ―where linguistic and cultural evidence reveals similarities of a lesser order but still sufficient to raise the presumption of historical connection…‖ Finally, one population from each sampling province of related cultures was chosen; the independence of each unit in terms of historical origin and cultural diffusion is maximal with respect to the other societies in the sample.30 Table 1 gives SCCS descriptive statistics. The SCCS is drawn evenly from all world regions, including Africa and the Pacific. All subsistence strategies are represented at the world level, but there are no pastoralists in either the Pacific or North America; no foragers in the Circum-Mediterranean; and relatively few agriculturalists in the Americas, especially North America.31 Environmental variables show the poor physical characteristics that some regions were dealt (Sub-Saharan Africa‘s high pathogen stress) but also indicate that development is not fully determined by geography (East Asia‘s equatorial climate, North America‘s low pathogen stress). Interestingly, all regions are within one-third of a standard deviation of worldwide average agricultural potential (determined by land slope, soil quality, and climate); the largest deviation is in Sub-Saharan Africa and it is towards better agricultural potential. I return to the importance of environmental and geographical determinates of economic development in Section 5.3 below. Sub-Saharan Africa has relatively large societies spread thinly while the Pacific has very dense societies that nonetheless are quite small. East Asia‘s dense populations are distributed evenly into smaller communities, while Circum-Mediterranean populations (also dense) are more often concentrated in a few large towns.32 This suggests that population-based measures are not appropriate proxies for income everywhere. Instead I construct an income proxy33 that directly

30

The observations that make up this cross-section of world populations were not all recorded the same year or even the same decade. By design the date of observation (focus year) is that of the earliest high quality ethnographic description; 85 percent of the observations are recorded between 1850 and 1965. Often ethnographic information is based on interviews with informants who describe historical practices. This contributes to the selection of a sample relatively free of the influence of colonization by European powers. 31 This does not mean there is no pastoralism in the Pacific or foraging in the Circum-Mediterranean, only that no society in those regions gets a majority of subsistence production from those activities. 32 Acemoglu, Johnson, and Robinson (2002) make a theoretical and empirical case for urbanization as a proxy for income. Ashraf and Galor (2008) develop a case for total population as an income proxy (in Malthusian eras). Maddison (2001, 2005) uses both urbanization and population densities to construct income-per-capita from the year

17

TABLE 1 — DESCRIPTIVE STATISTICS World

Sub-Saharan Africa

CircumMediterranean

East Asia

Insular Pacific

North America

South America

Number of societies

186

28

28

34

31

33

32

Agriculturalists (%)

39

57

50

53

42

12

25

Pastoralists (%)

6

4

21

12

0

0

3

Foragers (%)

31

11

0

21

23

70

53

Mixed (%)

24

29

29

15

35

18

19

Population density (median range, per square mile)

5 to 25

5 to 25

26 to 100

26 to 100

26 to 100

1 to 5

1 to 5

Total population (median range, thousands)

10 to 100

100 to 1000

100 to 1000

100 to 1000

1 to 10

1 to 10

1 to 10

Urbanization (mean size range, persons)

200 to 399

200 to 399

1000 to 5000

200 to 399

100 to 199

100 to 199

50 to 99

Climate (median Köppen Geiger type)

equatorial

equatorial

arid/ temperate

temperate/ equatorial

equatorial

temperate

equatorial

Pathogen stress (world percentile)

50th

85th

50th

50th

40th

10th

60th

Observation year (median)

1915

1920

1920

1930

1930

1870

1928

Note: The Insular Pacific includes societies from countries with East Asian elements that are not connected to the Asian landmass (Indonesia, Malaysia, Philippines, and Taiwan) as well as aboriginal societies from two Neo-European countries (Australia and New Zealand). The Circum-Mediterranean includes societies from Europe, the Middle East, and North Africa. Societies from Iran and countries east of Iran are included in East Asia. Russia is split between the Circum-Mediterranean (Russians) and East Asia (Chukchee, Gilyak, Yukaghir, and Samoyed). North America includes Mexico. For subsistence activity categories, a society is defined as being a type if more than 55 percent of subsistence production comes from that activity. Foragers must get more than 65 percent of subsistence production from any combination of hunting, gathering, and fishing. Mixed economies get less than 55 percent of subsistence from any one category. Pathogen stress is a cumulative index of the presence of 7 separate pathogens.

measures output by aggregating physical capital, administrative capacity, and financial market depth. Table 2 presents the index components and their range while Figure 4 presents a map marking the location of all SCCS societies with development proxy levels indicated by marker size.34 There are nearly 2000 variables recorded in the SCCS, from weaning practices to sources

0. Empirical results (discussed below) hold when either urbanization or total population are substituted for the development proxy described below. 33 The SCCS does not observe prices or consumption. 34 Another technology, boat building, was observed but not included in the development index in order not to handicap locations where water transport is not feasible. Including boat building in the development index while dropping those societies for which water transport is infeasible does not change results or rankings. The development index including boat building is nearly perfectly correlated (ρ = 0.98) with the index excluding boat building.

18

TABLE 2 — DEVELOPMENT PROXY COMPONENTS World

World

Median

Mean

motorized vehicles

human carriers

pack animals

unimproved trails

paved roads

unimproved trails

improved trails

none

public, ceremonial, military, or industrial

none

personal residence

potters, weavers,

potters, weavers,

potters, weavers,

Minimum

Maximum

Land transport: method

human carriers

Land transport: routes

Large or impressive structures

Craft specialization

none

and

or

or

metalwork

metalwork

metalwork

Writing and records

none

true writing and records

mnemonic devices

mnemonic devices

Money

none

true money

alien currency

domestic articles

friends, relatives

banks

friends, relatives

internal money

Credit source

Note: To create the development proxy, categorical values for these seven variables are summed. The number of categories for each variable are as follows: land transport: method, writing and records, and money: 5; land transport: routes and credit source: 4; large or impressive structures and craft specialization: 3. The minimum and maximum values of the index are 7 and 29.

of political authority. Though certain variables record beliefs (in gods, for example), the object of all observations is actual cultural practice rather than opinions, forecasts, or moral judgments. This distinction is important if the goal is to identify an effect of behavior on development. A more detailed discussion of specific cultural behaviors appears in Section 5. 4.2.

Genetic Data

For approximately 60 percent of SCCS societies I match genetic information from CavalliSforza, Menozzi, and Piazza (1994), which observes allele frequencies for nearly 2000 populations on every inhabited continent and several Pacific islands. From variant frequencies 19

FIGURE 4 – SCCS SOCIETIES and DEVELOPMENT

Note: Larger marker size indicates greater levels of development as measured by the proxy described in the Table 2 and accompanying text. Lighter marker shading indicates lower levels of expected genetic heterozygosity as described in Figure 5 and accompanying text in Section 4.2.

20

population-wide genetic heterozygosity (variety) can be calculated. With I loci (sites where variants occur), J possible alleles (variants), and Pj the population frequency of each possible allele, heterozygosity is equal to: .

(4)

The more frequently any particular allele j occurs, the less frequently other potential alleles occur and heterozygosity decreases. If all alleles occur with equal frequency, heterozygosity increases. Information from only a small number of loci for a limited number of SCCS societies is available. Using a result from population genetics, I also calculate for all SCCS societies expected heterozygosity from alleles at over 780 loci. The measure, which is based on the previously described linear decay of genetic heterozygosity as migratory distance from East Africa increases, summarizes the evolution of these migrations: they occurred serially, with genetically nonrepresentative population subsets, from a single origin. Figure 5 demonstrates this relationship for two samples: populations from the Human Genome Diversity Cell Line Panel used to confirm the relationship between heterozygosity and distance from East Africa (Ramachandran et al. 2005, Prugnolle, Manica, and Balloux 2005, Liu et al. 2006, Li et al. 2008), and a subsample of SCCS populations.35 The previous Figure 4 exhibits expected heterozygosities obtained from applying Ramachandran et al. (2005) slope coefficients (dark dashed line in Figure 5) to SCCS societies - lighter marker shading means less genetic heterozygosity.36 Expected neutral heterozygosity has attractive properties. It captures the minimum effect that genetic drift due to a serial founder effect has on variation in total genetic information and does not contain the portion produced by selection or post-migration gene flow between populations (Ramachandran et al. 2005). This same portion of demographic history (―isolation by distance‖) is what drives cultural mutation and permanence described in equation (2). It is also the portion of demographic history the SCCS sample selection mechanism attempts to capture. By averaging over

35

Intercepts vary due to the number of loci considered. There is one population common to both samples: Chinese at roughly 10500 kilometers from Addis Ababa. 36 The expected SCCS heterozygosity shown in Figure 4 is positively correlated with actual SCCS heterozygosity based on 3 loci (shown as light diamonds in Figure 5) with a ρ approximately equal to 0.75.

21

FIGURE 5 — EXPECTED HETEROZYGOSITY FOR TWO SAMPLES

slope = -6.5 x 10-

6

slope = -9.9 x 10-6

Note: Expected genetic heterozygosity (dashed lines) versus migratory distance from East Africa for two different samples. Dark circles are actual heterozygosities calculated over 783 loci for all populations in the Human Genome Diversity Cell Line Panel (HGDP-CEPH). Light diamonds are actual heterozygosities calculated over 3 loci for a subset of populations in the SCCS. The only common population is Chinese at approximately 10500 kilometers from Addis Ababa. Migratory waypoints are those used in Ramachandran et al. (2005). HDGP-CEPH data and figure are adapted from Ramachandran et al. (2005); SCCS data and figure are author‘s calculation. When applying HGDP-CEPH slope coefficients to all SCCS societies, expected heterozygosity decays as shown in the previous Figure 4, where lighter shading indicates less heterozygosity.

22

many loci that are non-coding,37 expected heterozygosity ensures that the signal from any one locus, including those possibly under selection, is muted. 5.

Empirical Results

5.1.

Specification

Table A.1 in the appendix lists unconditional correlations among regressors38 used in estimating the following equation:

Yj = λy + βCj + δyAj + τyEj + σyXj + εy,

(5)

where Y is the development proxy (see Table 2), C is a vector of cultural behaviors taken from the SCCS (see Table 3), A and E are vectors describing subsistence activity and environmental conditions, respectively, X is a vector of additional controls, and λ is a constant. Empirical estimation of β should explicitly incorporate the unobservable partial effect of

Yj on Cj (see equations (1) through (3)); without it, direct estimation may produce inconsistent coefficients. But when heterozygosity Gj satisfies the relevance assumption and exclusion restriction, a first-stage estimation of the reduced-form equation for Cj,

Cj = λc + ΨGj + δcAj + τcEj + σcXj + εc,

(6)

combined with a second-stage estimation of (5) using predicted values of culture from (6) in place of observed Cj will just identify β . I show that Gj satisfies the relevance assumption using standard statistical tests on Ψ; overidentification tests and regional empirical patterns suggest Gj is uncorrelated with εy and correctly excluded from equation (5). 5.2.

What Might Class, Inheritance Rights, and Game Complexity be Good For?

Table 3 lists cultural behaviors in C. More detail is given in Appendix A, but a convenient shorthand for the pathways between Table 3 behaviors and development is the following: class stratification coordinates the division of labor, games are education and inheritance rights are an

37

Number of poly-allelic loci (300+, 750+, 1000+) considered changes intercepts and slope coefficients, but not the general conclusions demonstrated in Figure 3. Number of loci used does not substantively change instrument relevance or two-stage least squares coefficients in Tables 4 through 9 upcoming. 38 There are significant correlations between subsistence activities, environment, and cultural behaviors. Agriculture is not always found in the most agriculturally productive environments and agriculturally productive land is frequently situated in equatorial climates with relatively high levels of rainfall and pathogen stress. Intuitively, these characteristics dovetail with the account of early sedentary agriculture in Diamond (1987), which argues it initially led to worse health, nutrition, and demographic outcomes.

23

TABLE 3 — CULTURAL TECHNOLOGIES Categories

Minimum

Maximum

World Median

World Mean elite - control over scarce resources distinguished from proletariat

5

absence among freemen

complex (social classes)

incipient wealth distinctions not crystallized

Inheritance rights

3

absence of rights in real or moveable property

presence of rights in real and moveable property

presence of rights in real and moveable property

presence of rights in real or moveable property

Game complexity

4

none of skill, chance, strategy

all of skill, chance, and strategy

skill & chance or skill & strategy

skill & chance or skill & strategy

Class stratification

Note: Each cultural behavior is recorded as an ordered categorical variable. For class stratification, higher categories represent more crystallized, complex, and widespread stratification; purely political and religious statuses and individual-level variation in "repute achieved through skill, valor, piety, or wisdom" are excluded. For inheritance rights, higher categories represent the presence of rights over more forms of property. Though SCCS records include property recipients, I code the variable to represent only the presence of rights. For game complexity, higher categories represent more complex contests; only contests with an outcome, i.e. a winner and a loser, are objects of observation.

early form of property rights. Class stratification produces division of labor and specialization by providing rules for the separation of the larger population into subpopulations based on culturally-specific determinations of superiority and inferiority. Regardless of whether values like honor or purity are eventually replaced by pecuniary success as the basis of class, the division into subpopulations, once crystallized, encourages the transmission of group-specific skills, behaviors, and information. Not only does this facilitate specialization and the decentralized coordination of information relevant to production, but also the formation of group identity and the creation of markets for the wares or symbols associated with groups, all of which promote economic development. Games and play behavior function much as formal education or research and development: they encourage cognitive development and human capital acquisition by providing a consequence-free environment in which experimentation, trial and error, and the spontaneous recombination of known quantities or methods can provide novel and better solutions to social problems. The SCCS observes game complexity and not time spent in play; the interaction 24

described suggests that higher levels of game complexity are analogous to higher levels of formal schooling. Inheritance rights are an early form of property rights giving the property owner the freedom to allocate his property once he can no longer use it personally. The SCCS observes rights in both portable and permanent property and while they are correlated, they are not collinear. The developmental response to secure property rights is familiar and will not be discussed at length here; further details and references are included in Appendix B. 5.3.

Instrumental Variables Results

5.3.1. Cultural Behaviors and Development Table 4 presents results from two-stage least squares (2SLS) estimation of equations (5) and (6) with both actual and expected genetic heterozygosity as instrumental variable candidates. With the first latent variable extracted from factor analysis of all Table 3 cultural behaviors39 as the dependent variable, columns (1) and (2) employ actual heterozygosity calculated over 1 locus as the instrument. Columns (3) and (4) use actual heterozygosity over two loci, and (5) and (6) use expected heterozygosity from 783 loci applied to SCCS populations (see section 4.2). Panel A presents second-stage results first for the unconditional correlation of development with cultural behavior and then with a complete set of subsistence production, environmental, climate, and focus year controls. Panel B presents the corresponding first-stage coefficients on heterozygosity and confirms that it satisfies the relevance assumption – only when sample size is severely reduced (column (4)) is the first-stage relationship weak. Panel C presents OLS coefficients from a direct estimation of equation (5) with controls the same as those in Panel A. In Table 4 and all subsequent specifications, sample standard errors are calculated using the Huber-White estimator which is robust to heteroskedasticity in residuals εy and εc.40 The developmental benefits of these cultural behaviors are substantial: from second-stage results in Panel A, a standard-deviation increase in the first factor leads to increases of between 39

The correlation between class and either inheritance rights or game complexity is approximately 0.45; that between inheritance rights and game complexity approximately 0.17. The first factor extracted from the three technologies leaves a substantial amount of variation in each unexplained. 40 Though the SCCS sampling frame goes some way toward producing relatively independent observations, there is still likely to be some autocorrelation in the cultural variables along geographic, historical, or ancestral distances. The Huber-White variance estimator is robust to such patterns.

25

TABLE 4 — IV REGRESSIONS OF DEVELOPMENT PROXY (1)

(2)

(3)

(4)

(5)

(6)

6.416*** (1.20)

5.721* (2.80)

5.529** (1.62)

10.979 (6.91)

3.324*** (0.97)

3.335* (1.41)

Panel A: IV Second stage First factor Agriculture

0.683 (0.41)

0.721* (0.32)

1.128 (0.89)

Foraging

0.128 (0.39)

-0.031 (0.27)

0.839 (1.18)

Nomadic

1.801 (2.40)

0.756 (1.41)

1.197 (5.62)

Food variability

-0.219 (1.54)

-0.864 (0.93)

-6.440 (5.56)

Pathogen stress

-0.375+ (0.21)

-0.256 (0.16)

-0.684 (0.44)

Agricultural potential

0.084 (0.16)

0.108 (0.14)

0.183 (0.42)

Rainfall

0.253 (0.31)

-0.192 (0.31)

-0.457 (1.66)

Climate?

no

yes

no

yes

no

yes

Focus year?

no

yes

no

yes

no

yes

Panel B: IV First stage

Instrument heterozygosity is based on: 1 locus, 3 alleles 2 loci, 7 alleles expected (783 loci)

Heterozygosity R2

2.277*** (0.51)

1.334* (0.51)

3.404** (1.07)

1.823 (1.29)

1.256*** (0.23)

1.056*** (0.24)

0.16

0.45

0.18

0.43

0.19

0.48

5.577*** (0.60)

3.954*** (0.72)

4.736*** (0.98)

3.828* (1.43)

4.728*** (0.54)

3.880*** (0.61)

0.36 39

0.62 36

0.42 115

0.61 108

Panel C: Ordinary Least Squares First factor

R2 0.53 0.70 Number of observations 68 63 Legend: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

Note: The dependent variable is the development proxy (see Table 2). ―First factor‖ is the first latent variable extracted from factor analysis of all Table 3 cultural behaviors. Panel A reports second stage results from IV regressions with all listed environmental and subsistence activity control variables, instrumenting for culture with genetic heterozygosity; Panel B reports corresponding first stage results for different heterozygosity measures (first stage coefficients on other excluded variables have been suppressed to save space); Panel C reports coefficients from an OLS regression of the dependent variable on the cultural variable plus the same controls listed in Panel A (OLS coefficients on those controls have been suppressed to save space). Robust standard errors are in parentheses. In regressions with climate dummies, the dummy for temperate climates is omitted. In regressions with focus year dummies, the dummy for focus years 1936-1960 is omitted. Only societies with focus years greater than 1750 are included.

26

50 and 85 percent of a standard deviation of development; a simultaneous one-unit or onestandard deviation increase in all three cultural technologies working through the first factor increases development by approximately 115 to 190 percent of a standard deviation. Current gaps in log GDP per capita of approximately 115 percent of a standard deviation exist between Mongolia and Turkey, Thailand and Ireland, or Colombia and Japan. Table 5 repeats specifications from Table 4 with an additive index41 created by summing scores on all Table 3 cultural behaviors in place of the first factor. Here again only when sample size is reduced in column (4) is the first-stage relationship between heterozygosity and culture weak. The developmental benefits of increases along any cultural dimension remain substantial: a one-standard deviation increase in the additive index leads to a development boost of 50 to 90 percent of a standard deviation. A simultaneous one-unit increase in all index variables increases development by 60 to 110 percent of a standard deviation.42 Table 6 repeats specifications from Tables 4 and 5 with discrete cultural behaviors in place of the cultural aggregates.43 It also includes OLS specifications with each discrete cultural behavior entered simultaneously. Within a 2SLS specification, an individual one category increase in class stratification, inheritance rights, or game complexity leads to an increase of 40, 60, and 70 percent of a standard deviation of development, respectively. Or, from the lowest levels of class, inheritance rights, and games to the highest, individual increases of roughly 1.6, 1.2, and 2.0 standard deviations of development. With all three technologies controlled for, OLS44 coefficients (Panel B) on class stratification, inheritance rights, and game complexity are reduced approximately 14, 46, and 48 percent from columns 2, 4, and 6 respectively, but each remains statistically significant at the 10 percent level or better.

41

Unlike the first factor, which summarizes variance common to all the cultural technologies, the additive index incorporates all variation along any dimension. 42 For either Table 4 or Table 5 specifications, restricting the expected heterozygosity sample (columns (5) or (6)) to those populations that also have observations on actual heterozygosity (the roughly 65 observations in columns (1) or (2)) does lead to small increases (10 to 20 percent) in second-stage cultural coefficient size. The difference is not statistically significant. There is approximately no change in coefficients or adjusted R2 in the first stage for the same restriction. 43 Results in Table 6 summarize a naïve statistical exercise: if expected heterozygosity is a valid instrument for class stratification, say, it cannot technically be so for inheritance rights because it will remain correlated with development (through class stratification) after inheritance rights are held constant. I thank David Levine and Chad Jones for pointing out this naivety. 44 Attempting to instrument for any two of these cultural strategies simultaneously produces weak first-stage results for one of them.

27

TABLE 5 — IV REGRESSIONS OF DEVELOPMENT PROXY (1)

(2)

(3)

(4)

(5)

(6)

2.039*** (0.38)

1.854* (0.91)

1.784** (0.53)

3.574 (2.33)

1.069*** (0.31)

1.070* (0.45)

Panel A: IV Second stage Additive index Agriculture

0.683 (0.41)

1.112 (0.90)

0.806** (0.28)

Foraging

0.141 (0.39)

0.840 (1.22)

-0.075 (0.23)

Nomadic

1.741 (2.40)

0.937 (5.75)

1.663 (1.54)

Food variability

-0.266 (1.55)

-6.728 (5.89)

-0.838 (0.82)

Pathogen stress

-0.391+ (0.21)

-0.739 (0.44)

-0.284* (0.14)

Agricultural potential

0.086 (0.16)

0.226 (0.43)

0.044 (0.11)

Rainfall

0.287 (0.31)

-0.439 (1.69)

0.061 (0.26)

Climate?

no

yes

no

yes

no

yes

Focus year?

no

yes

no

yes

no

yes

Panel B: IV First stage

Instrument heterozygosity is based on: 1 locus, 3 alleles 2 loci, 7 alleles expected (783 loci)

Heterozygosity R2

7.166*** (1.62)

4.117* (1.66)

10.552** (3.36)

5.600 (4.14)

3.905*** (0.71)

3.290*** (0.77)

0.16

0.45

0.18

0.43

0.19

0.48

1.757*** (0.19)

1.244*** (0.22)

1.513*** (0.31)

1.211* (0.45)

1.491*** (0.17)

1.207*** (0.19)

0.36 39

0.62 36

0.42 115

0.61 108

Panel C: Ordinary Least Squares Additive index

R2 0.53 0.70 Number of observations 68 63 Legend: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

Note: The dependent variable is the development proxy (see Table 2). ―Additive index‖ is the sum of scores on all Table 3 cultural behaviors. Panel A reports second stage results from IV regressions with all listed environmental and subsistence activity control variables, instrumenting for culture with genetic heterozygosity; Panel B reports corresponding first stage results for different heterozygosity measures (first stage coefficients on other excluded variables have been suppressed to save space); Panel C reports coefficients from an OLS regression of the dependent variable on the culture variable plus the same controls listed in Panel A (OLS coefficients on those controls have been suppressed to save space). Robust standard errors are in parentheses. In regressions with climate dummies, the dummy for temperate climates is omitted. In regressions with focus year dummies, the dummy for focus years 1936-1960 is omitted. Only societies with focus years greater than 1750 are included.

28

TABLE 6 — IV REGRESSIONS of DEVELOPMENT PROXY (1) Panel A: IV Second stage Culture Table 4 controls?

(2)

Class stratification

(3)

(4)

(5)

(6)

Cultural Technology Regressor is: Inheritance rights Game complexity

2.740*** (0.61)

2.076* (1.04)

3.036** (1.11)

3.117+ (1.67)

5.183*** (1.34)

3.575* (1.73)

no

yes

no

yes

no

yes

1.816*** (0.38)

1.101* (0.40)

1.195*** (0.23)

0.932*** (0.23)

1.067*** (0.22)

0.855** (0.29)

0.12

0.32

0.20

0.52

0.13

0.21

2.122*** (0.28)

1.595*** (0.27)

(7)

(8)

All independently

no

yes

1.706*** (0.35)

1.387*** (0.30)

1.551** (0.52)

1.114+ (0.59)

Panel B: IV First stage Expected heterozygosity R2 Panel C: Ordinary Least Squares Class stratification Inheritance rights

3.385*** (0.52)

2.071** (0.65)

Game complexity 0.36 0.55 R2 Number of observations 153 144 Legend: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

0.19 120

0.46 113

2.460*** (0.53)

1.593** (0.47)

0.816+ (0.50)

0.832+ (0.48)

0.13 145

0.47 136

0.42 115

0.60 108

Note: The dependent variable is the development proxy (see Table 2). Cultural technology regressors used are listed at the top of each column. Panel A reports the second stage results from IV regressions with the respective cultural variable and all listed environmental and subsistence activity control variables (coefficients on those controls have been suppressed to save space), instrumenting for each cultural technology with expected genetic heterozygosity. Instruments for all three cultural technologies are not available simultaneously. Panel B reports corresponding first stage results (first stage coefficients on other excluded variables have been suppressed to save space); Panel C reports coefficients from an OLS regression of the dependent variable on the respective cultural technology plus the same controls listed in Panel A (OLS coefficients on those controls have been suppressed to save space). Robust standard errors are in parentheses. In regressions with climate dummies, the dummy for temperate climates is omitted. In regressions with focus year dummies, the dummy for focus years 1936-1960 is omitted. Only societies with focus years greater than 1750 are included.

The dependent variable is not a standard measure of income and most SCCS societies are unfamiliar, so the economic meaning of a standard-deviation increase in development is not transparent. Comparing societies near the bottom of the distributions of all three cultural technologies and development to societies near the middle to societies near the top involves comparing, for example, the Yahgan (South American foragers) or Kung! (South African foragers) to Natchez (North American foragers/agriculturalist) or Somali (East African pastoralists) to the Irish (Western European mixed agriculturalists/pastoralists) or Chinese (Western Asian agriculturalists).

29

In Tables 4 through 6, OLS coefficients from Panel C are sometimes smaller than 2SLS coefficients from Panel A. This is consistent with attenuation in OLS coefficients from measurement error in the cultural variables, which do not capture all relevant behavioral variation. Since differences between OLS and 2SLS coefficients for behavioral aggregates are smaller,45 the first factor and the additive index may be capturing less noisy shares of ―developmental culture‖ than singular technologies in Table 6. Since it is a chronological as well as geographic summary of populations, heterozygosity may contain the omitted variable S (time since settlement – see equation (3)) which creates gradual improvements in technology from extensive trial and error. As discussed in Section 5.5.3 below, time since the introduction of technologies is a determinate of income at the national level. If this applies at the population level, 2SLS coefficients on cultural behaviors (Table 3) will increase if heterozygosity and S are correlated. In OLS specifications with cultural variables and heterozygosity entered simultaneously (results not presented), the heterozygosity coefficient is not significant and occasionally negative, which at a minimum indicates empirical difficulty in partialing out the choice of cultural technologies from the time necessary to make them effective. 5.3.2. Culture, Crops and Development Diamond (1997) argues, in contrast to the thesis of this paper, that long-run income differences can be traced to the particular features of settlement environments interacting with coadaptation and domestication of potent plant and animal species. The accidental development of higher net energy ―food packages‖ allowed societies to support more members; these demographic changes in turn led directly to higher group incomes. Table 7 provides an empirical amplification of these hypotheses by testing the effects of variation in animal and crop domesticates.46 The most potent of all domesticated plants and animals were the cereals (for their annual production of large, edible seeds) and bovines (for their edible fat, proteins, and milk, and hauling power). The SCCS records the principal crops (10 cereals, 6 roots or tubers, and 3 tree fruits) and domesticated animals (pigs, ovides, equines, reindeer, camel, and bovines) each

45

This is most clearly visible for specifications with more observations – columns (5) and (6) in Tables 4 or 5. I do not include information concerning the ease with which locally-developed food production technologies might have been transferred to neighboring populations, so Table 7 contains incomplete specifications of Diamond (1997) hypotheses. 46

30

TABLE 7 — IV REGRESSIONS: DEVELOPMENT and FOOD PACKAGES (1)

(2)

(3)

(4)

First factor

5.448+ (2.80)

5.725* (2.47)

3.559** (1.28)

3.221* (1.40)

Wheat/Barley

2.903 (2.19)

2.100 (1.97)

4.256* (1.76)

3.149+ (1.71)

Rice

3.091 (2.21)

Panel A: IV Second stage

0.326 (1.72)

Bovines Agriculture

0.552 (0.39)

Foraging

0.165 (0.37)

Panel B: IV First stage Heterozygosity R2

0.619 (0.46)

1.575 (1.12) 0.728* (0.28)

yes

0.797** (0.27)

0.110 (0.21) -0.076 (0.28)

Animal Husbandry Other Table 4 controls?

3.081* (1.47)

yes

-0.075 (0.21) yes

yes

Instrument heterozygosity is based on: 1 locus, 3 alleles expected (783 loci) 1.317* (0.57)

1.514** (0.48)

1.102*** (0.24)

1.034*** (0.24)

0.44

0.45

0.49

0.50

3.523*** (0.74)

3.674*** (0.72)

3.522*** (0.62)

3.496*** (0.57)

0.71 63

0.65 108

0.63 108

Panel C: Ordinary Least Squares First factor

R2 0.74 Number of observations 63 Legend: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

Note: The dependent variable is the development proxy (see Table 2). ―First factor‖ is the first latent variable extracted from factor analysis of all Table 3 cultural behaviors. Panel A reports second stage results from IV regressions with all listed environmental and subsistence activity control variables, instrumenting for culture with genetic heterozygosity. Panel A includes indicators for crops (Wheat/Barley and Rice) and animal domesticates (Bovines and Ovides/Equines), which are discussed in the accompanying text. Panel B reports corresponding first stage results (first stage coefficients on other excluded variables have been suppressed to save space); Panel C reports coefficients from an OLS regression of the dependent variable on culture plus the same controls listed in Panel A (OLS coefficients on those controls have been suppressed to save space). Robust standard errors are in parentheses. In regressions with climate dummies, the dummy for temperate climates is omitted. Only societies with focus years greater than 1750 are included.

31

society maintains. These indicators are included in Table 7, which presents 2SLS estimates from specifications otherwise similar to columns (2) and (6) of Table 4. Second-stage coefficients on culture are similar in economic significance to Table 4 coefficients even when indicators for both cereals and bovines are included.47 Controlling for multiple plant and animal domesticates simultaneously (results not presented) does not change cultural coefficient size or significance level appreciably. Most control variables have broadly the same sign and significance patterns in Table 7 as in Table 4 and cultural coefficients are similar in size, though less precisely estimated, when those societies not receiving any plant or animal domesticates are excluded (results not presented). Table 7 suggests that the distribution of plants and animals does affect development. But cultural technologies, also unpredictable and persistent, continue to have a statistically and economically significant effect on development when food packages are held constant. There is no indication that a geographic or environmental profile completely determines income (or culture). In other words, culture may be regarded as another source of windfall luck in long-run development. Its effects are not diminished by luck (or its lack) from other sources. 5.4.

Cultural Variation at Work

Are the developmental benefits accruing to these cultural technologies evident in pairwise comparisons? The Olmec of Mesoamerica developed rapidly from a simple village-agricultural lifestyle to a complex and highly productive civilization requiring management of prodigious amounts of labor and in turn benefitting from regionally-unknown outputs like hydraulic technologies. Further south in Peru, the Chavín likewise developed rapidly from simpler modes of organization. Meanwhile, even though societies in the area between Olmec and Chavín homelands were also organized into village-agricultural units; were in contact with Olmec and Chavín cultures; participated in the same technical traditions; and inhabited land similarly rich in climatic, altitudinal, and vegetational variety and natural resources, those intermediate societies did not develop either Olmec or Chavín versions, or any other version, of sophisticated

47

In results not presented, Table 7 specifications were repeated with both the additive cultural index and individual cultural technologies in place of the first factor. There were no substantial changes to coefficient size or significance (relative to Tables 5 or 6) for any cultural behaviors.

32

civilization. Both Coe (1968) and Willey (1962) suggest the Olmec and Chavín ―genius‖ is a result of cultural changes in class stratification and the division of labor.48 Within the SCCS universe consider the Amhara, Aymara, and Haitians: in different regions on opposite sides of the equator (East Africa, 13 degrees latitude; South America, -16 degrees latitude; and the Caribbean, 19 degrees latitude, respectively), they are all agriculturalists who have received some of the most potent cereals available.49 They are similar in all of the environmental variables in Tables 4 through 6. But the Aymara are a middle-income group while the Amhara and the Haitians are in the top ten percent of all SCCS societies and only the Amhara and Haitians have complex class stratification. In addition, there is evidence that when both Haitians and Aymara experimented with different forms of class stratification, output responded in the manner predicted by Table 4 through 6 coefficients (see Appendix B). The Turks, Koreans, and Mapuche are even more similarly situated geographically, productively, and in terms of environmental endowments (they are all rice or wheat growers at approximately 38 degrees absolute latitude). Like the Amhara, Aymara, and Haitians, these three societies all have inheritance rights and complex games. But the Mapuche did not develop high levels of class stratification while the Koreans and Turks did. The Mapuche are in the 80th percentile of the SCCS income distribution while the Koreans are in the 90th and the Turks in the 95th percentile, respectively. The Chiricahua Apache and the Pomo are both foragers living in Western North America (-110 and -123 degrees longitude, respectively). Neither society is well-off absolutely, but the Pomo are one of the richest North American foraging populations while the Chiricahua are tied for third poorest. Pomo have inheritance rights while the Chiricahua instead destroy virtually all property associated with a deceased member. The Kutenai, the most developed North American foragers (also living in the West at -117 degrees longitude), have likewise developed inheritance rights.50

48

Both also suggest the mutation that produced stratification was an increase in production devoted to the prevailing secular religion. Willey (1962) says ―It does us no good to deny the sudden mutation of creative change to the aborigines of America. It is no easier to explain elsewhere than it is here. What we are seeking is probably in New World soil, but genius must arise from preconditions which to our eyes do not foreshadow it.‖ 49 Each of the following comparisons is explored at greater length in Appendix B. The summary in Section 5.4 is meant only as brief sketch. 50 In environmental endowments the Pomo were slightly better endowed than the Chiricahua while the Kutenai are intermediate between Chiricahua and Pomo.

33

5.5.

Robustness

5.5.1. Overidentification Tests The validity of the candidate instrument is crucial: identification of the true partial effect of cultural behavior on development requires an instrument uncorrelated with the error εy in equation (5). If genetic heterozygosity affects development directly, then cultural coefficients may be capturing these effects and IV specifications will produce biased estimates. An empirical test of the exclusion restriction for genetic heterozygosity alone is not available because true errors εy are unobservable. An alternative, based on 2SLS coefficient comparisons when the endogenous variable is over-identified, is presented in Table 8. Overidentification tests indicate whether estimation with a smaller subset of instruments produces 2SLS coefficients that differ by more than expected sampling error from specifications with a larger set of instruments. Conditional on the assumption that at least one of the candidates from the larger set is correctly excluded, a failure to reject the null hypothesis that coefficients differ only by sampling error gives some confidence in the exogeneity of all candidates. Table 8 provides results from testing overidentifying restrictions with evil-eye belief (see Section 3.4 and Figure 2) as the additional instrument. The worldwide correlation between evil eye belief and development is small and positive (ρ = 0.16), but like heterozygosity (see below) there is regional variation that suggests there is no direct relationship. For Sub-Saharan Africa, East Asia, the Pacific, North and South America, correlations are -0.16, -0.14, 0.11, 0.18 and 0.46, respectively.51 If evil-eye belief affects output, it should be through association with other behaviors – similarity in evil-eye belief is a proxy for similarity in ―developmental culture‖. Panel C in Table 8 demonstrates that for this candidate instrument and expected or actual genetic heterozygosity, the overidentifying restrictions for cultural technologies are not rejected, meaning that first-stage coefficients do not differ markedly when a larger subset of instrument candidates is used.52 Panels B and A show that when the evil eye belief is used in place of heterozygosity first- and second-stage coefficients are not always precisely estimated.53 These results show no evidence that genetic heterozygosity has a direct effect on development and 51

There are only 4 South American societies observed after 1750 with evil eye belief. The positive correlation for South America is driven almost entirely by Haiti. In specifications similar to those in Tables 4 through 7, evil-eye belief is never statistically significant when any of the other cultural behaviors are held constant. 52 For game complexity no candidate instrument was found to be a good first-stage predictor, leading to weak instruments and potentially severe inconsistency in second-stage estimates. 53 Estimates of the statistics underlying the tests are heteroskedasticity-robust in both the first- and second-stage.

34

TABLE 8 — OVERIDENTIFICATION TESTS (1)

(2)

Panel A: IV Second stage First factor Culture Table 4 controls?

(3) (4) (5) (6) Cultural Technology Regressor is: Additive index Class stratification

Inheritance rights 6.005* (3.03) no

9.726 (10.25) yes

3.184*** (0.95) no

3.596+ (2.95) yes

(8)

4.466* (1.96) no

5.533 (4.25) yes

1.430* (0.63) no

0.419** (0.14)

0.176 (0.15)

1.307** (0.45)

0.548 (0.48)

0.787** (0.25)

0.400 (0.26)

0.328* (0.13)

0.120 (0.48)

0.07

0.40

0.07

0.38

0.06

0.37

0.05

0.50

Panel B: IV First stage

1.784 (1.38) yes

(7)

Instrument: Evil eye belief

Evil eye belief R2 Panel C: Overidentification Tests

Additional Excluded Variable: Heterozygosity (Expected or Actual)

p-value:

Expected Actual

{0.52} {0.45}

{0.70}

{0.52} {0.41}

{0.69}

{0.57} {0.78}

{0.41}

{0.25} {0.82}

{0.46}

Number of observations:

Expected Actual

115 65

108

115 65

108

153 79

144

120 66

113

Legend: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 Note: The dependent variable in columns (1)-(8) is the development proxy (see Table 2). Panel A reports the second stage results from IV regressions with the respective cultural variable. Production, environment, climate, and focus year comprise exactly the set of controls listed in Table 4 (first and second stage coefficients on controls have been suppressed to save space). Panel B reports corresponding first stage results for the instrumental variable candidate Evil-eye belief (see accompanying text). Panel C reports in braces the p-value for the null hypothesis that the coefficient on the cultural technology in the second stage (Panel A) is the same as when instrumented using the Panel B instrument plus genetic heterozygosity. Robust standard errors are in parentheses. In regressions with climate dummies, the dummy for temperate climates is omitted. In regressions with focus year dummies, focus years 19361960 is omitted. Only societies with focus years greater than 1750 are included.

suggest the exclusion of this variable from equation (5) is a valid assumption. 5.5.2. Regional Patterns in Expected Heterozygosity and Development There are regional irregularities providing additional evidence that heterozygosity is correctly excluded from equation (5). Intra-region or intra-subsistence activity correlations between cultural technologies and development mirror world-wide correlations: class stratification, property rights, and game complexity are always positively correlated with development. The regional correlations of heterozygosity and development do not. For Sub-Saharan Africa, the Circum-Mediterranean, East Asia, the Pacific, North America, and South America, unconditional correlations are roughly -0.5, -0.3, -0.1, 0.3, -0.2, and 0.4, respectively, for expected heterozygosity calculated over 780 loci. 35

This regional pattern would entail the following for a direct effect of neutral genetic variety on development: it was drawn down efficiently among the first migrants, but as the pool of neutral genetic material continued to shrink absolutely, carrying more, then less, then more of it became efficient in places further away. If the regular decay of neutral genetic material happens to be correlated with the population frequency of genetic information that is shown to directly affect development, it remains difficult to explain the regional heterozygositydevelopment correlation pattern above. The genetic information useful for development would occur less frequently in a regular and monotone manner (see Figure 4 or 5) while its effect on development would fluctuate depending on region.54 5.5.3. Alternative subsamples and alternative outcomes Table 9 presents 2SLS and OLS regression results from specifications similar to those in Table 6 columns (2), (4), and (6) over several different SCCS subsamples.55 In addition to the regular outcome variable in panels A and B, in panels C and D an alternative measure of development, urbanization, is specified for each subsample (including the ordinary Table 6 sample). Though urbanization is not a perfect income proxy56, it is familiar (see Acemoglu, Johnson, and Robinson (2002)) and intuitively approachable. The SCCS selection rules described earlier produce a set of ―independent as possible‖ observations. In addition, SCCS societies received identification numbers with the property that two societies closest in number were judged to be closest historically, ancestrally, and culturally (and often, but not necessarily, geographically). By using this numbering scheme to select every second society, the most similar pairs can be dropped in a methodical manner and the independence of each data point from all others increased. Two different seed numbers were used to select two non-overlapping subsets of SCCS societies: these are the ―Odd‖ and ―Even‖ 54

The heterozygosity-development and heterozygosity-culture regional patterns (not presented) also indicate that it is similarity in vertically-transmitted genetic information, rather than the genetic information itself, that proxies for similarity in vertically-transmitted cultural information. That is, in some regions higher cultural scores are associated with more genetic variety and in other regions with less. 55 In results not presented, alternative samples identical to those in Table 9 were used in Tables 4 and 5 specifications. Conclusions are similar to those discussed below: both the first factor and additive index remain statistically significant and economically meaningful across alternative samples and alternative outcomes. 56 The SCCS-wide correlation of the development proxy and urbanization is approximately 0.75, but this average hides significant regional variation: among the richer regions (East Asia and the Circum-Mediterranean), the correlation is approximately 0.87; among the Americas, the correlation is 0.66; and among Sub-Saharan Africa and the Pacific, the correlation is approximately 0.42. Sub-Saharan Africa is the 3rd richest region ranked by urbanization but 2nd poorest ranked by the development proxy.

36

TABLE 9 — IV REGRESSIONS: DIFFERENT SAMPLES AND DIFFERENT OUTCOMES (1) Cultural Technology: sample:

(2) (3) Class stratification

(4)

Ordinary

With ancients

Odd

No SubSah. Afr.

2.076* (1.04)

3.236** (1.18)

1.377 (1.01)

1.595*** (0.27) 0.54 144

1.773*** (0.26) 0.56 154

1.691* (0.76)

1.707* (0.70)

(5)

(6) (7) Inheritance rights

(8)

No Circ.Med. Dependent Variable is Development Proxy Ordinary

With ancients

Even

4.091** (1.36)

3.117+ (1.67)

3.985* (1.66)

4.305+ (2.42)

1.434*** (0.37) 0.60 74

1.734*** (0.32) 0.59 120

2.457* (1.09) 0.758*** (0.19) 0.64 60

(9)

(10) (11) Game complexity

(12)

Ordinary

With ancients

No agri.

No N. Amer.

3.363+ (1.70)

3.575* (1.73)

5.528* (2.64)

3.064+ (1.83)

3.599+ (1.96)

2.071** 2.411*** 1.959* 1.828** (0.65) (0.65) (0.93) (0.65) 0.46 0.49 0.36 96 113 118 54 0.45 Dependent Variable is Urbanization

1.593** (0.47) 0.47 136

1.558** (0.47) 0.47 144

1.483* (0.57) 0.48 90

1.589** (0.51) 0.50 108

2.223* (0.97)

1.973+ (1.09)

1.757+ (1.01)

0.521 (1.49)

1.373 (0.89)

3.173+ (1.80)

3.663 (2.48)

2.576* (1.17)

4.231 (2.92)

0.634*** (0.14) 0.59 105

0.851** (0.32) 0.50 101

0.906** (0.32) 0.53 106

0.279 (0.39) 0.42 52

0.839** (0.31) 0.48 83

0.831*** (0.22) 0.55 123

0.845*** (0.22) 0.57 130

0.938** (0.32) 0.59 75

0.870*** (0.25) 0.54 103

Panel A: IV Second Stage culture Panel B: Ordinary Least Squares culture Adj. R-square Obs Panel C: IV Second Stage culture Panel D: Ordinary Least Squares 0.569*** 0.647*** (0.12) (0.12) Adj. R-square 0.55 0.58 127 136 Obs Legend: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001 culture

Note: In Panels A and B, the dependent variable is the development proxy (see Table 2). In Panels C and D, the dependent variable is urbanization (see accompanying text). Cultural technology regressors used are listed at the top of each column. Panels A and C report second stage IV coefficients, and Panels B and D OLS coefficients, from regressing the dependent variable on the respective cultural variable and all production, environmental, climate, and focus year controls from Tables 4 through 6 (second stage and OLS coefficients on these controls have been suppressed to save space). The cultural technology is instrumented in Panels A and C with expected genetic heterozygosity. The "Ordinary" samples replicate those in Tables 4 through 6: all SCCS societies with focus years greater than 1750. The "With ancients" sample adds back SCCS societies with focus years of 1750 or earlier. The "Odd" and "Even" samples skip every other "Ordinary" sample SCCS society (starting with society 1 or 2, respectively) in order to achieve a subsample of societies more culturally, historically, and geographically independent (see accompanying text). The ―No agri.‖ sample excludes societies producing more than 56 percent of subsistence output from agriculture. The "No Sub-Sah. Afr.", "No Circ.-Med.", and "No N. Amer." samples drop all SCCS from Sub-Saharan Africa, the Circum-Mediterranean, and North America respectively. Robust standard errors are in parentheses. The dummy for temperate climates and for focus years 1936-1960 are always omitted.

37

samples in Table 9. The ―With ancients‖ samples add back in all of those SCCS societies with focus years of 1750 or earlier. Though only about 6 percent of all SCCS societies have focus years from this era, the set includes well-known cultures like the Babylonians, Hebrews, Romans, Khmer, Aztecs, and Incas. The other subsamples are achieved by dropping entire regions or production strategies from the ordinary Table 4 sample. For the ordinary sample, increasing class stratification, inheritance rights, and game complexity are associated with greater urbanization.57 Both OLS and IV coefficients on cultural technologies typically increase when older SCCS societies are included (columns (2), (6), and (10)). Choosing a culturally-, historically-, and ancestrally-diffuse sample (column (3) or (7)) does affect estimate precision and first-stage instrument fit (results not presented), but the overall picture remains the same. Similarly, dropping entire regions or production strategies (column (4), (8), (11), or (12)) often changes estimate size and precision, but the general conclusions remain unchanged. 5.5.4. Alternative controls and alternative behaviors There are alternative ways to code most of the control variables, but alternative codings do not change results substantially. For instance, production can be entered as 0/1 indicators for whether a majority of production comes from that activity. Both the pathogen stress and agricultural potential indices can be disaggregated and the components entered individually. Absolute latitude can be entered in place of Köppen-Geiger climate classifications. The time trend can be divided into more or fewer intervals. In results not presented, I try various combinations of these alternative codings and find no substantial change to results or conclusions. Appendix Table A.2 presents several specifications with regional environmental characterizations given by a measure of distance from the equator (absolute latitude) and dummy variables for continental regions. The local environmental controls included in Tables 4 through 9 more closely capture conditions under which production takes place; Table A.2 includes 57

Urbanization in the SCCS is recorded as an ordered categorical variable, but the underlying data is log-linear so coefficients between development proxy and urbanization specifications are not directly comparable. However, a one-unit change in class, inheritance, or games produces a 72, 84, or 135 percent of standard deviation change in urbanization categories (from a 2SLS specification). OLS results do not change if either ordered logit or interval regression models are specified: from a direct interval regression of equation (5) with log(urbanization) as the Yj, a one-unit change in class, inheritance, or games produces an approximately 35, 40, or 50 percent of standard deviation change in log(urbanization).

38

instead variables that are more familiar to researchers. Given the discussion in Section 5.5.2, it is not surprising that regional dummies soak up much of the useful variation in heterozygosity and consequently render it a weak first-stage instrument. However, the OLS coefficients are remarkably stable even when all Table 4 through 9 controls are included in addition to regional dummies and absolute latitude. When first-stage relationships are not weak, the second-stage IV coefficients are generally larger with regional dummies in place. Three variables germane to the evolution of culture and outupt are state history (Bockstette, Chanda, and Putterman 2002), years since transition to agriculture (Putterman 2008), and colonial identifiers. Nation-state resolution makes them noisy in the SCCS: they are mapped from just two values to over 30 North American societies, for example. In specifications including each of these regional variables one by one and otherwise similar to Tables 4 through 6, there is little change to either OLS or IV coefficients on cultural technologies (results not presented). State history and years since transition to agriculture are often statistically significant, which indicates that being surrounded by a relatively advanced state is beneficial for development. This is expected given the enhanced opportunities for trade and labor mobility and the regularization of economic activity that statehood brings.58 Indicators for British, French, Spanish, or mixed colonial heritage are rarely statistically significant, but the coefficient pattern is regular: there is a small positive development boost in states with British colonial heritage and a larger development penalty in states with French, Spanish, or mixed colonial heritage.59 Finally, an informal test of the power of the empirical framework is provided by using other cultural behaviors as endogenous regressors (results not presented). Consider descent, marriage transactions, postpartum sexual intercourse taboos, local political succession, religion, and physical separation of adolescent boys from female relatives: the first three of these do not have a statistically or economically significant economically correlation with development in a direct OLS estimation of equation (5). 2SLS specifications with actual or expected genetic heterozygosity as the excluded variable confirm these results, though only descent has a nonweak first-stage correlation with genetic heterozygosity.

58

See Bockstette, Chanda, and Putterman (2002) for a more detailed discussion of the beneficial effects of state antiquity on development. 59 94 percent of SCCS societies are located in states that were colonized by European nations (Britain, France, Spain, Portugal, the Netherlands, Belgium, Italy, or Russia).

39

The latter three behaviors have statistically significant OLS correlations from direct estimation, but even when genetic heterozygosity satisfies the relevance assumption in the first stage, as it does for religion and separation of boys, the 2SLS coefficients on these behaviors are not statistically distinguishable from zero; for separation of boys, the second stage coefficient is positive while the direct OLS coefficient is negative. Genetic heterozygosity does not predict variation in all distinct cultural behaviors once other correlates have been held constant. Even when it does, significant OLS coefficients are not always confirmed in a 2SLS framework. 6.

Conclusion

The role of culture in economic development is not difficult to unpack logically. Culture is a set of norms or standards of behavior that develop within and for social groups of individuals. Cultural behaviors cannot be isolated from social transactions, and economic exchange is a social transaction. Cultural rules create incentives by specifying which transactions are prohibited, which individuals can be party to allowable transactions, the expectations associated with any transaction (which vary by an individual‘s culturally-determined status), and the short- and longterm outcomes flowing from such transactions.60 They also directly produce human capital. While it may be easy to trace logically, empirical identification of the link between culture and development that is both robust and economically meaningful is not straightforward. The data assembled in this paper permit me to make sound empirical advances. I use a rich database of cultural norms the primary sources for which are close observation of actual behavior and which are measured at the community level, with an equal distribution of data points from all major world regions, to overcome cultural measurement issues.61 I use population-level records of genetic information in an instrumental variables strategy designed to overcome simultaneity in the joint determination of income and culture. This variation-in-variation could not have been optimized by group members during the generation of cultural technologies. Instead, the generation of genetic variety proceeded by a regular process that gradually removed potential varieties of genetic material as populations moved farther away from an origin in East Africa.

60

In all of these ways cultural norms maintain multiple sets of fixed prices by, for example, making the fixed price of an activity infinite for some and not others. 61 The decentralized nature of the cultural data gathering is worth mentioning twice. The absence of a central authority or overarching goal leaves observers free to record what is culturally and contextually meaningful.

40

The schematic model developed in this paper relies on a common source for the generation and persistence of cultural and genetic variety. The migration and arrival of humans in unfamiliar environments set in motion a series of population-level biological and behavioral changes. The results of these changes have been perpetuated across generations by the vertical transmission of both sets of information. Since the transmission of cultural information proceeds with less fidelity, any correlation between genetic and cultural variation should be biased towards zero. Empirical results indicate that cultural technologies have a significant impact on development even after controlling for a wide range of environmental and geographic variables, production techniques, and a time trend. Results hold for alternative measures of development and several alternative samples. Class stratification, inheritance rights, and game complexity all produce behavioral changes which are beneficial for long-run development. I have argued that stratification encourages the division of labor, specialization, and the dissemination of jobspecific skills, techniques, and technologies across generations. Inheritance rights are an early form of property rights. Game complexity measures the sophistication of informal education and encourages the development of cognitive skills, including flexibility in problem-solving. Development of more sophisticated versions of any of these cultural technologies can lead to cross-sectional increases of up to 115 percent of a standard deviation of development. Cultural rules are part of the larger institutional background in which economic transactions take place. While significant headway has been made in uncovering the response of income to parts of that background, more is needed on the interaction among and between the latter. If the supposition that institutional backgrounds are cohesive with their development following a path set down by the very first institutions is correct, there is always a danger that omitted institutional variables cloud the empirical picture. Further research on the interaction between formal institutions and culture is possible within the SCCS. Evidence at the country level is also necessary to confirm that society-level mechanisms are at work in larger amalgamations of people.

41

REFERENCES

Acemoglu, Daron, Simon Johnson, and James Robinson. 2001. ―The Colonial Origins of Comparative Development: An Empirical Investigation.‖ American Economic Review 91(5)(December): 1369-1401. ————. 2002. ―Reversal of Fortune: Geography and Institutions in the Making of the Modern World Income Distribution.‖ Quarterly Journal of Economics 117(4)(November): 12311294. Akerlof, George, and Rachel Kranton. 2005. ―Identity and the Economics of Organizations.‖ Journal of Economic Perspectives, 19(1): 9-32. Alesina, Alberto and Nicola Fuchs-Schündeln. 2005. ―Good-bye Lenin (or not?): The effect of Communism on people‘s preferences.‖ Working Paper, Harvard University. Ashraf, Quamrul and Oded Galor. 2008. ―Cultural Assimilation, Cultural Diffusion and the Origin of the Wealth of Nations.‖ Working Paper, Brown University. Bisin, Alberto and Thierry Verdier. 2001. ―The Economics of Cultural Transmission and the Dynamics of Preferences.‖ Journal of Economic Theory 97(2): 298-319. Bockstette, Valerie, Areendam Chanda, and Louis Putterman. 2002. ―States and Markets: The Advantage of an Early Start.‖ Journal of Economic Growth 7(4)(December): 347-369. Bowles, Samuel. 1998. ―Endogenous Preferences: The Cultural Consequences of Markets and other Economic Institutions.‖ Journal of Economic Literature 36(March): 75-111. Cavalli-Sforza, Luigi Luca and Marcus Feldman. 1981. Cultural Transmission and Evolution. Princeton, NJ: Princeton University Press. Cavalli-Sforza, Luigi Luca and Marcus Feldman. 2003. ―The application of molecular genetic approaches to the study of human evolution.‖ Nature Genetics 33: 266-275. Cavalli-Sforza, Luigi Luca, Paolo Menozzi, and Alberto Piazza. 1994. The History and Geography of Human Genes. Princeton, NJ: Princeton University Press. Coe, Michael. 1968. ―San Lorenzo and the Olmec Civilization.‖ In Dumbarton Oaks Conference on the Olmec, Elizabeth Benson ed. Pp. 41-71. Washington, DC: Dumbarton Oaks. Diamond, Jared. 1987. ―The Worst Mistake in the History of the Human Race.‖ Discover May: 64-66. Diamond, Jared. 1997. Guns, Germs, and Steel: The Fates of Human Societies. New York, NY: Norton. Durham, William. 1991. Coevolution: Genes, Culture, and Human Diversity. Stanford, CA: Stanford University Press. 42

Fernandez, Raquel and Alessandra Fogli. 2007. ―Culture: An Empirical Investigation of Beliefs, Work, and Fertility.‖ Working Paper, New York University. Fletcher, Jason and Steven Lehrer. 2008. ―Using the ‗Genetic Lottery‘ within Families to Examine the Causal Effect of Health on Academic Achievement.‖ Working Paper, Yale University. Galor, Oded and Omer Moav. 2007. ―The Neolithic Revolution and Contemporary Variations in Life Expectancy.‖ Working Paper, Brown University. Giuliano, Paola. 2007. ―Living Arrangements in Western Europe: Does Cultural Origin Matter?‖ Journal of the European Economic Association 5(5): 927-952. Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2008. ―Long Term Persistence.‖ Working Paper 14728, National Bureau of Economic Research. Grimm, Michael, and Stephan Klasen. 2008. ―Geography vs. Institutions at the Village Level.‖ Working Paper 2259, CESifo. Hall, Robert and Charles Jones. 1999. ―Why Do Some Countries Produce So Much More Output per Worker than Others?‖ Quarterly Journal of Economics(February): 83-116. Harpending, Henry and Alan Rogers. 2000. ―Genetic Perspectives on Human Origins and Differentiation.‖ Annual Review of Genomics and Human Genetics 1: 361-385. Knack, Stephen and Philip Keefer. 1997. ―Does Social Capital Have an Economic Payoff? A Cross-Country Investigation.‖ Quarterly Journal of Economics (November): 1251-1288. Lahr, Marta Mirazon and Robert Foley. 1994. ―Multiple Dispersals and Modern Human Origins.‖ Evolutionary Anthropology 3(2): 48-60. Lahr, Marta Mirazon and Robert Foley. 1998. ―Towards a Theory of Modern Human Origins: Geography, Demography, and Diversity in Recent Human Evolution.‖ Yearbook of Physical Anthropology 41: 137-176. Li, Jun Z., Devin Absher, Hua Tang, Audrey Southwick, Amanda Casto, Sohini Ramachandran, Howard Cann, Gregory Barsh, Marcus Feldman, Luigi L. Cavalli-Sforza, Richard Myers. 2008. ―Worldwide Human Relationships Inferred from Genome-Wide Patterns of Variation.‖ Science 319: 1100-1104. Licht, Amir, Chanan Goldsmith, and Shalom H. Schwartz. 2007. ―Culture Rules: The foundations of the rule of law and other norms of governance.‖ Journal of Comparative Economics 35: 659-688. Liu, Hua, Franck Prugnolle, Andrea Manica, and François Balloux. 2006. ―A Geographically Explicit Genetic Model of Worldwide Human-Settlement History.‖ The American Journal of Human Genetics 79: 230-237. 43

Logan, Trevon, and Paul Rhode. 2008. ―Moveable Feasts: A New Approach to Endogenizing Tastes.‖ Working Paper, Ohio State University. Maddison, Angus. 2001. The World Economy: A Millenial Perspective. Paris: OECD. Maddison, Angus. 2005. ―Measuring and Interpreting World Economic Performance 1500-200.‖ Review of Income and Wealth 51(1). Mellars, Paul. 2006. ―Why did modern human populations disperse from Africa ca. 60,000 years ago? A new model.‖ Proceedings of the National Academy of Sciences 130(25): 93819386. Munshi, Kaivan and Nicholas Wilson. 2008. ―Identity, Parochial Institutions, and Occupational Choice: Linking the Past to the Present in the American Midwest.‖ Working Paper 13717, National Bureau of Economic Research. Murdock, George. 1967. Ethnographic Atlas: A Summary. Pittsburgh, PA: The University of Pittsburgh Press. Murdock, George and Douglas White. 1969. ―Standard Cross-Cultural Sample.‖ Ethnology 9: 329:369. Nunn, Nathan. 2008. ―The Long-Term Effects of Africa‘s Slave Trades.‖ Quarterly Journal of Economics 123(1). Olson, Mancur. 1965. The Logic of Collective Action. Cambridge, MA: Harvard University Press. Olsson, Ola, and Douglas Hibbs. 2005, ―Biogeography and long-run economic development,‖ European Economic Review 49: 909-38. Prugnolle, Franck, Andrea Manica, and François Balloux. 2005. ―Geography predicts neutral genetic diversity of human populations.‖ Current Biology 15(5): R159-R160. Putterman, Louis. 2008. ―Agriculture, Diffusion, and Development: Ripple Effects of the Neolithic Revolution.‖ Economica 75: 729-748. Ramachandran, Sohini, Omkar Deshpande, Charles Roseman, Noah Rosenberg, Marcus Feldman, and Luigi Luca Cavalli-Sforza. 2005. ―Support from the relationship of genetic and geographic distance in human populations for a serial founder effect originating in Africa.‖ Proceedings of the National Academy of Sciences 120(44): 15942-15947. Roland, Gerard. 2004. ―Understanding institutional change: Fast-moving and slow-moving institutions.‖ Studies in Comparative Economic Development 38(4): 109-131. Roland, Gerard and Jon Jellema. 2006. ―Institutional Clusters.‖ In What Drives Growth in the Transition Countries?: Institutions, Markets, and Economic Performance. European Bank for Reconstruction and Development. 44

Smith, Adam. 1759 [1966]. The Theory of Moral Sentiments. New York: A.M. Kelley. Spolaore, Enrico, and Romain Wacziarg. 2009. ―The Diffusion of Development.‖ Quarterly Journal of Economics, forthcoming in vol. 124(2), May. Tabellini, Guido. 2007. ―Institutions and Culture.‖ Presidential Lecture, European Economic Association Meetings, Budapest. Temple, Jonathan and P. Johnson. 1998. ―Social Capability and Economic Growth.‖ Quarterly Journal of Economics (August): 965-990. Weber, Max. 1905 [2001]. The Protestant Ethic and the Spirit of Capitalism. London: Routledge Classic. Willey, Gordon. 1962. ―The Early Great Styles and the Rise of the Pre-Columbian Civilizations.‖ American Anthropologist 64(1): 1-14.

45

A.

Appendix: Regressor Correlations and Alternative Geographic Controls

TABLE A.1 — UNCONDITIONAL CORRELATIONS AMONG REGRESSORS Class stratification

Inheritance rights

Game complexity

Agriculture

Pastoralism

Foraging

Nomadic

Food variability

Pathogen stress

Agricultural potential

Rainfall

Inheritance rights

0.45

Game complexity

0.44

0.17

Agriculture

0.35

0.61

0.07

Pastoralism

0.28

0.25

0.14

-0.05

Foraging

-0.43

-0.63

-0.13

-0.73

-0.55

Nomadic

-0.35

-0.52

-0.06

-0.77

0.13

0.52

Food variability

0.23

0.26

0.33

0.37

-0.07

-0.23

-0.36

Pathogen stress

0.12

0.31

-0.11

0.45

0.11

-0.41

-0.30

-0.02

Agricultural potential

0.13

0.18

-0.02

0.39

-0.14

-0.19

-0.35

0.12

0.36

Rainfall

-0.05

0.13

-0.27

0.30

-0.32

-0.02

-0.38

-0.03

0.40

0.40

Equatorial climate

-0.18

0.06

-0.32

0.27

-0.21

-0.12

-0.24

-0.20

0.49

0.33

0.57

Arid climate

0.01

-0.06

0.13

-0.18

0.35

-0.08

0.28

0.02

-0.04

-0.20

-0.61

Temperate climate

0.21

0.04

0.23

-0.10

-0.05

0.15

-0.02

0.24

-0.44

-0.12

-0.09

Note: Class stratification, inheritance rights, and game complexity are all ordered categorical variables with 5, 4, and 3 categories, respectively. Higher categories mean more of the cultural behavior as explained in the text. Production variables agriculture, pastoralism, and foraging are measured as percent of subsistence coming from that activity. Nomadic, food variability and climate variables are categorical 0/1 indicators; rainfall is an ordered categorical variable. Pathogen stress is a cumulative index of the presence of 7 separate pathogens. Agricultural potential aggregates indicators for slope, soil, and climate. Rainfall is an ordered categorical variable measuring average annual rainfall in millimeters.

46

TABLE A.2 — IV REGRESSIONS: ALTERNATIVE REGIONAL GEOGRAPHIC CONTROLS

Geographic controls:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

AL

SSA

SSA, CM

SSA, CM, EA

SSA, CM, EA, PAC

SSA, CM, EA, PAC, NA

SSA, EA, NA, SA

AL, SSA, CM

AL, EA, PAC

AL, NA, SA

AL, SSA, CM, +

Dependent Variable is Development Proxy

Panel A: IV Second Stage First factor (n=114)

2.774* (1.22)

6.845*** (1.36)

10.097** (3.10)

8.333* (3.80)

12.649+ (7.57)

0.471 (11.45)

6.635** (2.38)

10.934* (4.22)

2.648* (1.02)

-1.096 (3.98)

10.315** (3.91)

Class stratification (n=152)

1.944* (0.86)

4.388*** (1.06)

6.906* (2.79)

9.523 (8.14)

27.749 (73.70)

-1.170 (9.98)

3.802** (1.40)

7.294+ (4.05)

1.739* (0.71)

0.211 (2.06)

6.023* (2.70)

Inheritance rights (n=119)

3.262+ (1.93)

9.285** (2.76)

11.388* (4.85)

5.517 (3.86)

9.913 (7.68)

7.105 (30.07)

23.245 (24.99)

10.302* (4.82)

2.969* (1.43)

-13.523 (42.79)

10.263+ (5.49)

Game complexity (n=144)

2.390* (1.09)

8.159*** (2.21)

9.386* (3.67)

5.972* (2.53)

4.943** (1.60)

-1.580 (7.72)

8.252* (3.87)

14.283 (10.32)

2.183* (1.02)

0.307 (1.44)

25.361 (34.99)

0.988*** (0.21)

1.180*** (0.26)

0.865* (0.34)

0.856+ (0.50)

0.689* (0.32)

1.200*** (0.21)

0.617* (0.31)

0.734* (0.32)

0.47

0.40

0.41

0.41

0.41

0.42

0.40

0.48

0.49

0.48

0.51

1.194** (0.40)

1.852*** (0.46)

1.164* (0.55)

0.764 (0.82)

0.301 (0.85)

-0.735 (1.60)

1.951* (0.75)

0.852 (0.54)

1.399*** (0.41)

0.827 (0.67)

1.060+ (0.55)

0.29

0.27

0.29

0.29

0.29

0.29

0.28

0.31

0.29

0.30

0.35

0.708*** (0.20)

0.878*** (0.25)

0.752* (0.35)

1.156* (0.46)

0.698 (0.46)

-0.259 (0.61)

0.271 (0.28)

0.713+ (0.36)

0.973*** (0.21)

0.094 (0.24)

0.737+ (0.37)

0.50

0.50

0.50

0.50

0.52

0.53

0.54

0.50

0.54

0.53

0.53

1.124*** (0.25)

0.999** (0.31)

0.901* (0.38)

1.300* (0.51)

1.800*** (0.51)

-0.918 (1.02)

0.852+ (0.48)

0.470 (0.36)

1.195*** (0.26)

1.278*** (0.34)

0.275 (0.39)

0.22

0.11

0.11

0.11

0.14

0.21

0.19

0.26

0.24

0.23

0.28

Panel B: IV First Stage First factor R2 Class stratification R2 Inheritance rights R2 Game complexity R2

Instrument: Genetic Heterozygosity 0.621 -0.595 0.940* (0.49) (0.88) (0.38)

47

TABLE A.2 — CONTINUED

Geographic controls:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

AL

SSA

SSA, CM

SSA, CM, EA

SSA, CM, EA, PAC

SSA, CM, EA, PAC, NA

SSA, EA, NA, SA

AL, SSA, CM

AL, EA, PAC

AL, NA, SA

AL, SSA, CM, +

3.774*** (0.54)

4.597*** (0.49)

4.721*** (0.52)

4.486*** (0.56)

4.570*** (0.56)

4.378*** (0.58)

4.497*** (0.55)

4.337*** (0.57)

3.852*** (0.54)

3.715*** (0.61)

3.915*** (0.62)

1.550*** (0.24)

1.790*** (0.24)

1.732*** (0.26)

1.636*** (0.27)

1.660*** (0.27)

0.60 - 0.68 1.610*** 1.730*** (0.26) (0.24)

1.526*** (0.26)

1.549*** (0.23)

1.515*** (0.26)

1.376*** (0.27)

2.099** (0.71)

2.025** (0.61)

1.997** (0.63)

1.538* (0.70)

1.911** (0.58)

2.475*** (0.45)

2.104*** (0.47)

1.706*** (0.47)

1.351** (0.44)

1.804*** (0.47)

Panel C: Ordinary Least Squares First factor (n=114) R2 Class stratification (n=152) R2 Inheritance rights (n=119)

0.54 - 0.62 2.013** (0.61)

2.589*** (0.59)

2.354*** (0.62)

2.112** (0.64)

2.261** (0.70)

1.484*** (0.43)

2.708*** (0.40)

2.606*** (0.40)

2.476*** (0.40)

2.508*** (0.43)

R2 Game complexity (n=144)

1.886* (0.73) 0.44 - 0.57

R2 Legend: + p<0.1; * p<0.05; ** p<0.01; *** p<0.001

2.269*** (0.46) 0.48 - 0.58

Note: The dependent variable is the development proxy (see Table 2). Geographic controls included are listed at the top of each column, where AL = absolute latitude, SSA = Sub-Saharan Africa, CM = Circum-Mediterranean, EA = East Asia, PAC = Insular Pacific, NA = North America, and SA = South America. SSA, CM, EA, PAC, NA, and SA are all indicator variables. Cultural technology regressors used are listed at the left in rows. First factor is the first latent variable extracted from a factor analysis of the three cultural technologies Class stratification, Inheritance rights, and Game complexity (see Table 3). Panel A reports the second stage results from IV regressions with the row cultural variable, the column geographic controls, and subsistence activity and focus year variables from Table 4 (second stage coefficients on geographic, subsistence, and focus year variables have been suppressed to save space). Column 11 includes additional controls: Wheat/Barley and Bovines (see Table 5), as well as Food variability, Pathogen stress, Agricultural potential, Rainfall, and Climate (see Table 4). The number of observations, (n=...), applies to columns 1 through 10; Column 11 specifications lose 6, 10, 6, and 8 observations for the First factor, Class stratification, Inheritance rights, and Game complexity, respectively. Each row's cultural technology is instrumented with expected genetic heterozygosity. Panel B reports corresponding first stage results (first stage coefficients on other excluded variables have been suppressed to save space); Panel C reports coefficients from an OLS regression of the dependent variable on the respective cultural technology plus the same controls listed in Panel A (OLS coefficients on those controls have been suppressed to save space). Robust standard errors are in parentheses. In regressions with focus year dummies, the dummy for focus years 1936-1960 is omitted. Only societies with focus years greater than 1750 are included.

48

B.

Appendix: Cultural Technologies and Development Pathways

B.1.

Inheritance Rights

The theoretical and empirical association of property rights and income has been studied previously. Costly investments will not be made if the embodied investment can be seized by others without recompense (Demsetz 1967 and Alchian and Demsetz 1973). Inheritance rights over movable and fixed property are recorded in the SCCS, so ―threat of expropriation‖ is not precisely the correct description of the mechanism creating disincentives to invest. Inheritance rights give property holders the right to dispose (post-mortem) of possessions according to how they see fit rather than protect property from being seized. Besley (1995) identifies three separate channels via which property rights and investment could be linked. Property rights insecurity may act like a random confiscatory tax on possessions, property rights might make possessions more easily collateralizeable (leading to lower interest rate charges), or property rights could lead to expanded trading opportunities and the ability to exploit gains from trade, resulting in enhanced investment incentives by expanding the range of profitable investments. Specifically in the case of inheritance, property rights may also encourage the adoption of longer time horizons (the life cycle of a lineage, for example, instead the life cycle of an individual) and may lessen the likelihood of a tragedy of the commons. Besley (1995) finds property rights predictive of investment in agricultural production in modern Ghana. Bandiera (2007) finds qualitatively similar conclusions for investment in agricultural production in modern Nicaragua. Brunt (2007) finds property rights increase agricultural output through fixed capital investment in South Africa; British administration finally offered this novel institution near the middle of the 19th century. Johnson, McMillan, and Woodruff (2002) find that property rights are an important predictor of reinvestment in manufacturing firms in a sample of formerly communist countries. B.2.

Competitive Contests

Time spent in play has been described by anthropologists and other social scientists as human capital acquisition. Play can be thought of as an educational program or pure research and development activity with all the attendant externalities from either (Gosso et al. 2005).

49

What follows is a partial list of functions or consequences of play in both human and closely related species: language acquisition, learning to deceive and recognize deception, brain growth and cognitive development, motor skill development, behavioral flexibility, the development of social strategies, including the management of conflict and compromise, an originator and disseminator of novel behavior, which in turn can furnish new food sources, change social positions, extend geographical range, and open new ecological niches (Fagen 1981); a discoverer of roles or identities, general problem solving, creativity, divergent thinking (Bock 2005); mental flexibility and ―planfulness‖, a discoverer of the benefits of delayed gratification, adjudication, control, and leadership (Power 2000); and a preparer for adult competency in subsistence work (Bock and Johnson 2003). Equally important as the skill correlates of play is the manner or method in which play teaches and in particular the incentive and reward structure implicit in play. All the works on play cited in this appendix mention that play is a behavior that lacks consequences. This combined with the reward structure from play, in which a winner takes all but losers remain nearly as well off as they were before materially, creates incentives for experimentation. Bateson (2005) suggests play as a nearly cost-free optimization search technique, encouraging innovation and creativity in moving away from the previously established optimum. The SCCS does not observe or record time spent in play, but rather the complexity of competitive contests. This assists in empirical identification of the society-wide developmental impact of play as the observations will not be affected by a latent labor-leisure tradeoff dimension that could be heterogeneous across levels of development. In addition, by focusing on the elements of contests rather than the type of play, the observations do not implicitly penalize lower income societies for a lack of, e.g., fantasy or imaginative play (Gosso et al. 2005). Empirical evidence suggests that those children who show the highest level of play involvement and complexity score highest on cognitive functioning tests (Power 2000). Empirical tests in this paper show the same mechanisms may be working at the society-wide level. If play is a proxy for informal education or research and development, then such results should be expected: some of the increased productivity formal schooling produces comes from externalities created by socialization and preference formation not explicitly part of the curriculum. Also like a formal educational program, complexity in play can lead directly to increased productivity by encouraging innovation along more than one vector.

50

B.3.

Class Stratification Within groups of humans living together, inequality or a ―pecking order‖ may be

inevitable, since humans belong to a species of mammal that shows dominance behavior (Hofstede 2001). Or, maybe the dominance, command, and obedience required to sustain a social ordering, together with a desire to evaluate, creates inequality everywhere (Béteille 1977). Perhaps instead inequality arises everywhere because a society needs to motivate people to fill positions in the social structure, and some positions require special talents or training that is costly (Davis and Moore 1945). SCCS observations show that inequality and social hierarchies may not exist in every human society. Where they do occur, however, they occur with variability in complexity and fixedness. Not observed in the SCCS but also relevant for the relationship between stratification and development are the values concerning the exercise of power that can be expected to accompany a system of power differentiation. Since ―history is full of dead and overthrown stratification systems‖ (Smith 1966), any link between variation in stratification systems and development is perforce a link between legitimacy or longevity of stratification systems and development. Stratification systems vary also by the attributes to which social power is distributed or from which power flows. French and Raven (1959) classify the bases of social power into five types: reward power, coercive power, legitimate power (based on rules), referent power (based on personal charisma), and expert (specialist) power. The SCCS observation of status excludes purely political or religious statuses as well as individual-level variation in repute achieved through skill, valor, piety, or wisdom, but there is otherwise no precise indication of which attributes produce social strata nor why those attributes and not others are evaluated. The channel from more complex and durable stratification to increased development is that of division of labor and specialization.62 Some stratification may be sufficient for a stable social ordering and a stable social ordering is necessary before production and exchange can occur.63 But further durability and complexity in stratification leads to rewards being distributed and power exercised through a ranking of occupations (Hodge et al. 1966) and pecuniary success 62

Orans (1966) provides an early argument for such a channel. Social stratification may not be necessary for order in large groups: the Chimbu of New Guinea, where exchange is virtually all reciprocal and interpersonal relations are regulated by an elaborate ritual system, have population densities reaching 400 people per square mile, yet have no kings, chiefs, social stratification, or ranking. Tellingly, they also have ―none of the trappings of civilization whatsoever‖ (Flannery 1972). 63

51

can displace other values such as honor or purity as the basis for evaluation (Béteille 1977). The highest level of class stratification observed in the SCCS is ―Complex (social classes) – correlated in large measure with extensive differentiation of occupational classes.‖ A division of labor by class in turn makes possible all the gains from specialization and exchange. A division of labor that exists across generations may facilitate the transfer of occupationspecific skills and promote further productivity within occupations and class. The argument is not that inequality and dominance must necessarily precede the division of labor, specialization, and gains from trade. Instead, social stratification serves as a focal point or a coordination mechanism that eventually produces a division of labor even if it was not developed consciously for such a project. In other words, a non-contingent class structure can eventually produce an ordered and durable division of labor and specialization, no matter what the original basis of differentiation was. Furthermore, neither class stratification nor subsequent division of labor are irreversible: Carneiro (1967) provides an example of such a system created and disassembled annually among North American Plains Indians. Each band was relatively egalitarian during most of the year, but during the annual buffalo hunt, the aggregation of several different bands into a single hunting group was followed by the creation of a ruling council, a paramount chief with considerable authority, men‘s societies that assisted the ruling bodies, a police to enforce the rules that prevailed during the hunt, and a regulatory body to ―[preserve] order on the march and during the sun dance.‖ The degree of mobility between classes is also important as a moderator of the effects of class division. The more rigid is the stratification system, the less chance a society will have of discovering any new facts about the talent of its members (Tumin 1953). The incentives created by the division of people into categories could be blunted by the maintenance of an unyielding class structure. There is recently a recognition within economics that non-monetary incentives like status, role, and identity can enhance performance (Besley and Ghatak 2008; Akerlof and Kranton 2005). Belonging to a class may not only enhance productivity directly through specialization but indirectly by providing motivation to ascend to the next higher class or to cement one‘s status in one‘s own class. Bernard and Killworth (1973) provide early anecdotal evidence that such incentives are important for motivating labor. There may be other positive effects of stratification on development. For example, Béteille (1977) notes that stratification and differential status creates markets for most things associated with status. Durkheim (2006 [1893]) claims that effective regulation of any 52

profession comes primarily from members, who are the only actors with the information necessary to determine costs and benefits. He also claims that when men with common interests seek each other out, this group will transcend the individual and become the basis for a generalized morality. Bernard and Killworth (1973) provide evidence that two uniquely specialized groups will produce a greater joint product if each is held to the standards of their own profession, even if production takes place under great intergroup friction (when standards vary across professions). Flannery (1972) stresses the greater information-processing potential of specialized groups when that information is federated (across ranks) but not centrally organized. Thus, social stratification may function much like the Hayek (1945) price mechanism: though each group knows only a fraction of the collective information set, resource allocations can still be based on local conditions and opportunities. Garicano (2000) translates the Hayek price mechanism into a hierarchical production-knowledge allocation mechanism that facilitates problem-solving at the firm level, but the efficiency-enhancing aspects of such hierarchies apply to any group engaged in production. C.

Appendix: Cultural Technologies at Work

Consider the Amhara and the Aymara. Both are agriculturalists, the Amhara cultivating teff (a cereal) in a temperate climate at approximately 13 degrees north of the equator in East Africa, the Aymara cultivating barley in an equatorial climate at approximately 16 degrees south of the equator in South America.64 Both societies face similar levels of pathogen stress, seasonal food variability, and soil quality. But the Amhara are approximately twice as developed as the Aymara.65 Though they show similar levels of game complexity and property rights, the Amhara have the highest level of class stratification and the Aymara the second-lowest. Hoben (1973) notes that the Amhara are composed of a number of distinct segments differentiated from one another by ―occupation, power, and honor‖, and the division between peasants and elite is compounded by a distinction between layman and churchmen, and between military and non-military. Messing (1957) documents that there is an ethnic division of skilled labor and endogamy within the ethnic/occupation groups. Though an aura of ―untouchability‖ 64

Amhara secondary crops include barley, wheat, maize, millet, and hops. Aymara secondary crops include potatoes and quinoa. 65 Though the SCCS places the Amhara in the Circum-Mediterranean region (within which they have mean levels of development), they would be the most highly developed Sub-Saharan African society. They are near the 85th percentile in the SCCS-wide distribution of development. In terms of development, the Aymara are a mean South American society, and near the 45th percentile in the SCCS-wide distribution.

53

applies to certain occupational groups (smiths, tanners) and practitioners are often suspected of witchcraft and black magic, nonetheless their wares are ―much prized‖ not only for their material utility but also because of their ―supernatural‖ strength, endurance, and workmanship. Consequently, ethnic/occupational groups have monopolies on their respective markets as any manufactures by outsiders are thought to be substandard.66 Within the Aymara, however, social classes have not arisen: though there are leaders, leadership does not carry with it privilege or special rewards for families. On the social plane all are considered equal; what classes exist are extremely fluid and mobile, lack formal internal organization, and fluctuate with the fortunes of families and individuals. Neither is there a division of labor: everyone is a farmer even if income is supplemented by learning a trade Furthermore, the means of acquiring wealth are extremely limited and available to few. (Tschopik 1951).67 Compare both to Haitians, agriculturalists producing the cereal sorghum in an equatorial climate 19 degrees north of the equator in Central America. They face similar levels of pathogen stress, rainfall, food variability, and soil quality as the Amhara and the Aymara. The Haitians were as developed as the Amhara and had an equal level of class stratification (and are similar to both Amhara and Aymara in inheritance rights and game complexity). Though two separate episodes of colonization brought physical and human capital, in the roughly 100 years between the departure of the first colonial power and the arrival of the second, there was variation in the presence and ubiquity of class stratification. Leyburn (1941) notes that only during that century when class divisions (primarily between political, military/police, farming, and artisanal groups) were mandated and enforced from above and through endogamy – in essence, a re-creation of the French colonial hierarchy without the French colonials – did Haitian economic production reach colonial levels.68

66

Importantly, Messing (1957) also notes that wealth enables each of the many social classes to achieve some upward mobility and Hoben (1970) notes that differences between classes were largely of ―achieved rank, not ‗blood‘, ideals, or basic style of life. Peasant and [elite] shared similar military aspirations, the same ancestors, and above all the same vertical principles of social organization and ranking.‖ 67 Under the Inca aristocracy, Aymara society was more internally stratified, and there were several levels of class in between the Inca elite and Aymara commoners. During the same era, Aymara cultural, social, and political complexity was ―considerably enriched.‖ (Tschopik 1946) 68 Furthermore, when those class divisions were essentially erased by an ―in perpetuity‖ re-distribution of land, formal recognition of private property, the encouragement of small-scale, personal agricultural production, and the creation of a rudimentary social safety net (employer-provided medical care, social security for the elderly), output fell steadily and did not recover (Leyburn 1941).

54

Koreans, Turks, and the Mapuche of Chile are all agriculturalists growing primarily cereals (rice or wheat) in temperate climates at approximately 38 degrees absolute latitude. They face similar levels of pathogen stress, food variability, and soil quality, and have similar levels of game complexity and property rights. However, the Koreans and the Turks have developed high levels of class stratification while the Mapuche are relatively egalitarian; Turks are in the 95th, Koreans the 90th, and Mapuche the 80th percentile of SCCS-wide development. Within the Mapuche there are ―loose‖ divisions between wealthy and commoners, but no organization into occupational groups (though many different crafts are practiced) and personal rank and prestige derive chiefly from ―martial prowess and from wealth; generous hospitality, and eloquence in speech‖ (Cooper 1946). Furthermore, no great authority attaches to leaders; they serve mostly consultative or persuasive roles with no authority to sanction or coerce. (Faron 1968). Tasks that require more labor than is available to the household or kin group are organized communally on a voluntary basis, and the work parties that form for land clearing, housebuilding, or road repair do not remain once the work is finished. Similarly, much day-today work is organized on a reciprocal basis rather than through market exchange. Though individual ownership of land is well-established and agricultural partnerships both within the Mapuche and with neighboring Chileans are frequent, sophisticated, and economically productive, still little renting or selling of land occurs because it amounts to a loss of ―cultural integrity and agrarian status‖ (Faron 1961). Finally, consider two foraging societies, the Pomo and the Chiricahua, settled at 39 and 32 degrees latitude (respectively) in the Western United States. The Pomo are semisedentary and face no variability in food supply while the Chiricahua are nomadic and face some variability in food supply. Neither has high levels of pathogen stress. Among all foragers, the Pomo are relatively well developed (90th percentile) while the Chiricahua are just below average (40th percentile). Among all societies in the SCCS, the Pomo are still well off (75th percentile) and the Chiricahua are not (25th percentile). Though they have identical levels of game complexity, the Pomo have a higher level of inheritance rights and class stratification. Gifford (1923), Barrett (1952), and Loeb (1926) all document that the Pomo had definite rules about individual ownership of economically productive assets like trees, dams, gathering grounds, and fishing landings. These and other productive capital like harpoons, bows and arrows, fish and duck nets, or dwellings were passed to family members on the death of their original owner (Loeb 1926). Opler (1965) notes that the Chiricahua instead destroy all property 55

belonging to, any objects regularly used by, and any gifts coming from the deceased. So powerfully do they wish to completely alter the situation associated with the deceased that camp life is completely reconstructed in a new locality and the deceased‘s home is usually burned. The pre-death settlement is rarely visited. Loeb (1926) also notes that there is hereditary occupational specialization among the Pomo, especially for chieftainship, membership in the secret society, the office of doctor or shaman, hunting, and fishing. Importantly for the Pomo, a person can be a member of a profession only if he is ―initiated into the profession by an elder relative and given the magical outfit and charms necessary‖; outfits, charms, and work songs are all private property handed down within families. Among the Chiricahua, however, distinctions of hereditary status are recognized, but not rigidly maintained; ability and personal magnetism are as important as birth and wealth for determining leadership. Furthermore, there is virtually no specialization in occupation: every hunter also fishes and makes his own arrows, for example (Opler 1965).

56

APPENDIX REFERENCES

Akerlof, George, and Rachel Kranton. 2005. ―Identity and the Economics of Organizations.‖ Journal of Economic Perspectives, 19(1): 9-32. Alchian, Armen, and Harold Demsetz. 1973. ―The Property Rights Paradigm.‖ Journal of Economic History 33: 16-27. Bandiera, Oriana. 2007. ―Land Tenure, Investment Incentives, and the Choice of Techniques: Evidence from Nicaragua.‖ World Bank Economic Review 21(3): 487-508. Barrett, Samuel. 1952. ―Material Aspects of Pomo Culture.‖ Bulletins of the Public Museum of the City of Milwaukee 20. Bateson, Patrick. 2005. ―The Role of Play in the Evolution of Great Apes and Humans.‖ In The Nature of Play: Great Apes and Humans, Anthony Pellegrini and Peter Smith, eds. Pp 13-26. New York: The Guilford Press. Bernard, H. Russel, and Peter Killworth. 1973. ―On the Social Structure of an Ocean-Going Rsearch Vessel and Other Important Things.‖ Social Science Research (2): 145-184. Besley, Timothy. 1995. ―Property Rights and Investment Incentives: Theory and Evidence from Ghana.‖ Journal of Political Economy 103(5): 903-937. Besley, Timothy, and Maitreesh Ghatak. 2008. ―Status Incentives.‖ American Economic Review 98(2): 206-211. Béteille, André. 1977. Inequality among Men. Oxford: Basil Blackwell. Bock, John. 2005. ―Farming, Foraging, and Children‘s Play in the Okavango Delta, Botswana.‖ In The Nature of Play: Great Apes and Humans, Anthony Pellegrini and Peter Smith, eds. Pp 254-284. New York: The Guilford Press. Bock, John, and Sara Johnson. 2003. ―Subsistence Ecology and Play Among the Okavango Delta Peoples of Botswana.‖ Human Nature 15(1): 63-81. Brunt, Liam. 2007. ―Property Rights and Economic Growth: Evidence from a Natural Experiment.‖ Centre for Economic Policy Research Discussion Paper 6404. Carneiro, Robert. 1967. ―On the Relationship Between Size of Population and Complexity of Social Organization.‖ Southwestern Journal of Anthropology 23(3): 234-243. Cooper, John. 1946. ―The Araucanians.‖ Bulletins of the Bureau of American Ethnology 143(ii): 687-760. Davis, Kingsley, and Wilbert Moore. 1945. ―Some Principles of Stratification.‖ American Sociological Review 10: 242-249. Demestz, Harold. 1967. ―Toward a Theory of Property Rights.‖ American Economic Review 57: 347-359. 57

Durkheim, Emile. 1893 [2006]. ―The Division of Labor in Society.‖ In Inequality: Classic Readings in Race, Class, and Gender, David Grusky and Szonja Szelényi, eds. Pp. 55-64. USA: Westview Press. Easterly, William. 2007. ―Inequality does cause underdevelopment: Insights from a new instrument.‖ Journal of Development Economics 84: 755-776. Fallers, Lloyd. 1964. ―Social Stratification and Economic Processes in Africa.‖ In Economic Transition in Africa, Melville Herskovits and Mitchell Harwitz, eds. Pp. 113-130. Evanston: Northwestern University Press. Fagen, Robert. 1981. Animal Play Behavior. New York: Oxford University Press. Faron, Louis. 1968. Mapuche Indians of Chile. New York: Holt, Reinhart and Winston. Faron, Louis. 1961. ―Mapuche Social Structure: Institutional Reintegration in a Patrilineal Society of Central Chile.‖ Illinois Studies in Anthropology 1. Flannery, Kent. 1972. ―The Cultural Evolution of Civilizations.‖ Annual Review of Ecology and Systematics 3: 399-246. French, John and Bertram Raven. 1959. ―The Bases of Social Power.‖ In Studies in Social Power, Dorwin Cartwright, ed. Pp. 150-167. Ann Arbor: University of Michigan, Institute for Social Research. Garicano, Luis. 2000. ―Hierarchies and the Organization of Knowledge in Production.‖ Journal of Political Economy 108(51): 874-904. Gifford, Edward. 1923. Pomo Land on Clear Lakes. Berkeley, CA: University of California Press. Gosso, Yumi, Emma Otta, Maria de Lima Salum e Morais, Fernando Jose Leite Ribeiro, and Vera Silvia Raad Bussab. 2005. ―Play in Hunter-Gatherer Society.‖ In The Nature of Play: Great Apes and Humans, Anthony Pellegrini and Peter Smith, eds. Pp 213-253. New York: The Guilford Press. Hayek, Friedrich. 1945. ―The Use of Knowledge in Society.‖ American Economic Review 35(4): 519-530. Hoben, Allan. 1970. ―Social Stratification in Traditional Amhara Society.‖ In Social Stratification in Africa, Arthur Tuden and Leonard Plotnicov, eds. Pp. 187-224. New York: The Free Press. Hoben, Allan. 1973. Land Tenure among the Amhara of Ethiopia: the dynamics of cognatic descent. Chicago: University of Chicago Press. Hofstede, Geert. 2001. Culture’s Consequences: Comparing Values, Behaviors, Institutions, and Organizations Across Nations. Thousand Oaks: Sage Publications.

58

Hodge, Robert, Donald Treiman, and Peter Rossi. 1966. ―A Comparative Study of Occupational Prestige.‖ In Class, Status and Power: Social Stratification in Comparative Perspective, Second Edition, Reinhard Bendix and Seymour Lipset, eds. Pp. 309-321. New York: The Free Press. Johnson, Simon, John McMillan, and Christopher Woodruff. 2002. ―Property Rights and Finance.‖ American Economic Review : 1335-1356. Lancy, David. 1996. Playing on the Mother Ground: Cultural Routines for Chldren’s Development. New York: The Guilford Press. Leyburn, James. 1941. The Haitian People. Westport, CT: Greenwood Publishers. Loeb, Edwin. 1926. Pomo Folkways. Berkeley, CA: University of California Press. Messing, Simon. 1957. The Highland-Plateau Amhara of Ethiopia. Ph.D. dissertation. University of Pennsylvania. Power, Thomas. 2000. Play and Exploration in Children and Animals. Mahwah, New Jersey: Lawrence Erlbaum and Associates. Pellegrini, Anthony, and Peter Smith. 2005. ―Play in Great Apes and Humans.‖ In The Nature of Play: Great Apes and Humans, Anthony Pellegrini and Peter Smith, eds. Pp 3-12 . New York: The Guilford Press. Opler, Morris. 1941 [1965]. An Apache Life-Way. New York, NY: Cooper Square Publishers Orans, Martin. 1966. ―Surplus.‖ Human Organization 20: 24-32. Smith, M.G. 1966. ―Pre-Industrial Stratification Systems.‖ In Social Structure and Mobility in Economic Development, Neil Smelser and Seymour Lipset, eds. Pp. 141-176. Chicago: Aldine Publishing Company. Tschopik, Harry. 1946. ―The Aymara.‖ Bulletins of the Bureau of American Ethnology 143(ii): 501-573. Tschopik, Harry. 1951. ―The Aymara of Chcuito, Peru.‖ Anthropolgoical Papers of the Museum of Natural History 44: 137-308. Tumin, Melvin. 1953. ―Some Principles of Stratification: A Critical Analysis,‖ American Sociological Review 18: 387-394.

59