Admixture in the Americas

Can racial ancestry predict cognitive ability and socioeconomic outcomes?

Overview

• Background
• Review of between-individual results
• New between-region results
• New between-country results

Background

• (Additive) Genetic hypothesis for ethnoracial differences → trait level should follow admixture%
• Three levels of analysis:
1. Within country, within region
2. Within country, between region
3. Between countries
• We will look at all three in this talk

Within country and region

• Two different ways of studying within country, within region:

​

• Group-level: Sort individuals into groups based on ancestry, calculate group differences

​

• Individual-level: Measure admixture in individuals, correlate with trait level

Group-level studies

• Literature mostly mentions only a few, often old, contested findings:
• Proponents cite: Minnesota Transracial Adoption Study (Weinberg et al, 1992) and various old studies cited in Shuey (1966)

​

• Critics cite: Black conscripts-German mothers in post-WW2 Germany (Eyferth, 1961), small institutional studies (Tizard, 1974; Moore, 1986), elite sample studies (Witty and Jenkins, 1934, 1936)

​

• Lack of attention to newer studies and meta-analyses

​

Review of ethnoracial differences in the Americas

-- More recent data for 13 countries: Bolivia, Brazil, Canada, Chile, Colombia, Costa Rica, Ecuador, Guatemala, México, Nicaragua, Paraguay, Peru, U.S.

-- Generally agrees with Lynn’s (2008) findings (some results shown below)

Review of biracial studies (European-African) : meta-analysis

• Weighted means: .38 and .44, about 50/50 as expected by genetic hypothesis

J Fuerst, E Kirkegaard - Admixture in the Americas: part 1 (in progress)

Transracial adoption studies

• Commonly discussed: Minnesota Transracial Adoption Study (Weinberg et al, 1992), small Korean adoption study: Frydman & Lynn (1989), small African and mixed race adoption study Tizard (1974)
• There are much larger newer studies not mentioned: Odenstad et al (2008)

​

Transracial adoption studies 2

• “There were 5,942 individuals in the adoptee study group: 3,237 individuals were born in the Far East (2,658 were born in South Korea), 1,422 in South Asia, 871 in Latin America, and 412 in Africa.”

Lindblad et al 2003

http://emilkirkegaard.dk/en/?p=4995

Individual-level: New data 1

• Review of studies published between 2004 and 2014
• 46 studies using educational and SES indexes: 23 US, 4 Mexico, 3 Brazil, 3 Chile, 3 Colombia, 2 Puerto Rico, 2 Peru, 4 others
• 100% gave results in the expected direction
• 40 (87%) with p<.05
• Effect sizes typically not given, so could not meta-analyze
• In Latin America, ethnic identity appears to not be a good explanation of outcome variance net of genotype (e.g., Ruiz-Linares et al.’s (2014) five country study)

New data 2

• One study with direct IQ measurement (Picture Vocabulary) and genomic ancestry: large dataset (PING), same as recent “poverty shrinks the brain” study (Noble et al, 2015)

​

Review of older admixture studies (European-African)

• Proxy measures of racial ancestry: skin color, blood groups, nose width, lip thickness + IQ

Color/Reported Ancestry and IQ (in African Americans)

Looked at the ADD Health, NLSY97, and National longitudinal Study of Freshman surveys

--replicated earlier findings (regarding color and parental/self reported ancestry)

--evidence of a Jensen Effect on the color association

Table: Color, parental reported European ancestry, and Table: Jensen Effect on color-IQ correlation

aptitude in the NLSY97 in the African American population

Between regions and countries: the data

• Mexico, Brazil and American countries
• Data gathered from various studies reporting genomic admixture by region within Brazil and Mexico and for countries. Representative studies sought.
• For US, used census data for ethnoracial groups in conjunction with reported group admixture %

Admixture within Mexico 1

Admixture within Mexico: Cognitive ability and European%

Admixture within Mexico: Cognitive ability and MR prediction

Admixture within Mexico: S factor and European%

• S factor = general socioeconomic factor extracted from 21 variables, see: http://emilkirkegaard.dk/en/?p=5032
• S x HDI2010 = .95

Admixture within Mexico: S factor and MR prediction

Admixture within Brazil 1

• Data points overlap because reliable estimates were only available by region, not state.

​

• Available state level data (n=16) shows the same effect

Admixture within Brazil: Cognitive ability

​

​

​

Admixture within Brazil: Cognitive ability (more results)

​

Regression results for PISA 2012 using 16 state admixture values computed from Mouta et al.’s (2015) meta-analysis:

​

​

​

​

Beta values ~ same with missing state data filled in using regional data

Admixture within Brazil: S factor

• S factor = general socioeconomic factor extracted from 21 variables, see: http://emilkirkegaard.dk/en/?p=5154
• S x HDI2010 = .94

Admixture within the US 1

Admixture within the US: Cognitive ability

Admixture within the US: S factor

• S factor = general socioeconomic factor extracted from 25 variables, see: http://emilkirkegaard.dk/en/?p=4801
• S x HDI2010 = .87

Admixture between sovereign states

Validation (National Genomic Ancestry)

Validated National Genomic Ancestry Estimates (European, African, Amerindian)

​

-- National reflectance data

-- Average CIA % Ethnoracial Identity

-- Putterman and Weil’s World Migration Matrix data set

​

Table: European national genomic, ethnoracial self-identification, migration, and reflectance correlations

​

​

​

​

Admixture between sovereign nations: Cognitive ability

Admixture between sovereign nations: S factor

• S factor = general socioeconomic factor extracted from 54+42 variables, see: http://openpsych.net/ODP/2014/09/the-international-general-socioeconomic-factor-factor-analyzing-international-rankings/
• S x HDI2013 = .97

Admixture between sovereign nations: HDI2013

• HDI is better here than the more precise S factor because it has more datapoints: 25 vs. 35

Admixture between states and nations

States and nations : Cognitive ability

States and nations : HDI

Note: We use HDI here because of the problems with re-scaling the S factor scores to the same level

Let’s try modeling!

• Traditional MR modeling: choose some predictors + a dependent variable
• Only try and show a few of the possible models
• Possibility of researcher degree of freedom
• So we simply try and report all of them, sort by R^2 adjusted
• Similar procedure for Bayesian linear regression (using BayesFactor package for R)

http://emilkirkegaard.dk/en/?p=4881

General modeling procedure

• Independents: Use two genomic predictors
• Independents: Climatic data, parasite load (national level), tourists (national level)
• Cognitive ability, S factor/HDI as dependents
• Find the top 5 models for each meta-modeling

​

National Analysis: Background

• Different explanations for inter-national differences offered:
• Van de Vliert (2013) → effects of contemporaneous climate
• Eppig, Fincher, and Thornhill (2010) → effects of contemporaneous parasite load
• Lynn (2008) → effects of historic climate etc. by way of race
• Lynn (2012) → suggested tourism to explain West Indian wealth (e.g., Barbados)
• Interested in the independent effect of racial ancestry.

National Analysis: Caveats

​

• Independents covary
• Racial ancestry covaries with climate, parasite load, and other independents
• Causal pathways are tangled
• E.g., parasite load is partially a consequence of climate and of (cognitive differences associated with) racial ancestry. It is also partially a cause of the distribution of racial ancestry (i.e., West Africans imported to more parasite infested regions because of genetic resistance).

Sovereign nations: Cognitive ability

Frequentist top5

​

Bayesian top5

Sovereign nations: S factor

Frequentist top5

​

Bayesian top5

Sovereign nations: HDI

Frequentist top5

​

Bayesian top5

Mexican states: Cognitive ability

Frequentist top5

​

Bayesian top5

Odd, African ancestry has a positive beta? (Estimates were unreliable -- uncorrelated between studies)

Mexican states: S factor

Frequentist top5

​

Bayesian top5

Mexican states: HDI

Frequentist top5

​

Bayesian top5

US states: Cognitive ability

Frequentist top5

​

Bayesian top5

US states: S factor

Frequentist top5

Bayesian top5

Bizarre result!? Too many predictors? Suppressor predictors?

US states: HDI

Frequentist top5

​

Bayesian top5

International Colorism?

Cognitive ability / S / HDI

Skin color

Discrimination

Cognitive ability / S / HDI

Skin color

Genomic ancestry

Genomic ancestry

Colorism model

Genetic model

Modeling shows that (on the national level) models with skin color are not parsimonious. The best model is Eugenomic alone (frequentist and Bayesian modeling agree).

Can predict but not well when Eugenomic is there too

Future Avenues of Research

Collected numerous additional variables:

​

Literacy: e.g., Age 15 literacy, mean years of schooling, early 20th century numeracy indexes

Social well being indexes: e.g., GDP and Life expectancy

Crime/STD: e.g., Violent crime and HIV rates

Other cognitive measures: e.g., GRE, TOEFL, English Proficiency Index, and 2014 National IQs (still under construction)

Scientific Competencies: e.g., Per capita scientific papers published and per capita scientific researchers

Allows for testing of other hypotheses e.g., relative impact of economic policies/institutions vs. ancestry on GDP/growth.

​

Specific questions:

Is racial ancestry associated with crime rates on the national level (as suggested by Rushton and Templer (2009))?

Does Caribbean tourism moderate the GDP-ancestry association (as suggested by Lynn and Vanhanen (2012))?

​

​

​

​

​

​

​

Future Avenues of Research: Example

Inter-national homicide rate by ancestry

​

Beta SE CI.lower CI.upper

​

Afrgenome 0.65 0.21 0.21 1.09

Amergenome 0.54 0.20 0.12 0.95

​

​

​

​

​

​

​

Data set is freely available!