From Behavioral Genetics to Economics

Gene, Environment, Human Capital, and Inequality: From Behavioral Genetics to Economics

Junjian Yi and James Best

Econ 350, Winter 2012 This draft, March 5, 2012

Junjian Yi and James Best () Gene and Environment 1 / 57 Introduction Nature, Nurture and Economics

The nature-nurture debate has been over the heritability of many diﬀerent features of human behaviour: Cognitive capacities Personality Tastes Habits Health Ideology This can be seen, with some reservations, as a debate about the heritability of: Capacities Preferences

Junjian Yi and James Best () Gene and Environment 2 / 57 Introduction Nature, Nurture and Economics

An individuals capacities are a part of an agents budget constraint. Therefore, the relative importance of heritability for an individual’s budget is determined by two factors: 1 The level of heritability for particular capacities. 2 The relative importance of those capacities to the budget constraint. These two factors along with the heritability of preferences are very important for informing: The analysis of heterogeneity across individuals. The impact of policy on individual capacities and/or preferences.

Junjian Yi and James Best () Gene and Environment 3 / 57 Introduction The Findings of Behavior Genetics

“The nature-nurture debate is over. The bottom line is that every thing is heritable...” -Three Laws of Behavior Genetics and What They Mean (Turkheimer, 2000) The Three “Laws” 1 All human behavioral traits are heritable. Intelligence Personality Tastes Ideology 2 The effect of being raised in the same family is small than the effect of genes. Much of the BG literature finds very small family effects. 3 A substantial portion of the variation in complex human behavioral traits is not accounted for by the effects of genes or families.

Junjian Yi and James Best () Gene and Environment 4 / 57 Introduction Diﬀerent Approaches

Variance Decomposition Twin Studies Adoption Studies Parent Child Correlations Adoption Studies Twin Parent Studies Genetic Bio-markers

Junjian Yi and James Best () Gene and Environment 5 / 57 Genetics Lamarck and Weissman

Lamarck postulated that the environment/activities of the parents could affect the genes that they pass on to their children. August Weissman’s (1834-1914) germ plasm theory states: Germ cells (sperm and eggs) are the sole determinants of inheritance. That the germ cells produce somatic cells and are unaffected by somatic cells: genotype affects phenotype but not vice-versa. This has been seen as a rejection of Lamarkian evolution The Weismann barrier has largely been vindicated by 20th Century genetic research: DNA copies to RNA which determines phenotype, RNA only very very rarely affects DNA.

Junjian Yi and James Best () Gene and Environment 6 / 57 Genetics Mendel’s Laws (Mendel, 1866)

Mendel’s laws are fundamental to quantitative genetic analysis. Law of Segregation (First Law) Each individual has a pair of alleles (an allele is a form of a particular gene-type) and they pass one randomly selected allele in each pair to their oﬀspring. In the case that the two received alleles are diﬀerent one will be dominant, will be expressed in the phenotype, and the other recessive, will not be expressed in the phenotype. Law of Independent Assortment (Second Law) The probability that an allele is copied to a gamete (sperm or egg) is independent of which other genes have been copied to that gamete.

Junjian Yi and James Best () Gene and Environment 7 / 57 Genetics Mendelian Inheritance

Junjian Yi and James Best () Gene and Environment 8 / 57 Genetics Genetic Relatedness

Mendels laws and the Weissman barrier imply a particular genetic relatedness across families.

Junjian Yi and James Best () Gene and Environment 9 / 57 Genetics Mendels Laws, Additiviy and Normality

The number of a particular allele of a set of genes is binomially distributed. If those genes are additive then the phenotype will also be binomially distributed.

As the number of genes relating to the characteristic of interest increase then the characteristic will approximate a normal distribution.

Junjian Yi and James Best () Gene and Environment 10 / 57 Genetics Advances in Genetics and Evolutionary Theory

Beyond Mendel’s Laws (Smith & Ebrahim 2003) Linkage disequilibrium: the probability of receiving some genes is connected to the probability of receiving others. Epistasis, Pleiotropism and Genetic heterogeneity: These are different ways in which gene affects are non-additive. The above phenomena affect the genetic relatedness of family members and the normality of phenotypic distribution.

Junjian Yi and James Best () Gene and Environment 11 / 57 Genetics Advances in Genetics and Evolutionary Theory

The study of Epigenetics (Charness 2011) has yielded evidence that: There is non-genetic transmission of phenotypic characteristics both through the germ line and the soma. Environment can induce changes in cells that allow for non-genetic transmission of characteristics. Epigenetics has implications for the aﬀect of genotypic relatedness of family members on phenotypic similarities. It implies the possibility that the Weissman barrier is permeable: Implying something resembling Lamarkian evolution is also possible. CAUTION: To my limited knowledge the signiﬁcance of epigenetic processes relative to genetic processes is not clear.

Junjian Yi and James Best () Gene and Environment 12 / 57 Univariate ACE Model The Univariate ACE Model: Basic Setup

The ACE model is used to estimate what proportion of variance in some observable outcome is accounted for by genes, by shared environment and by non-shared environment. To do this it makes an additivity assumption: Assumption I: Each factor aﬀects the outcome additively. This gives the following basic set-up:

Yi = aAi + cCi + eEi (1)

Y : outcome. A: Additive genetics C: Common environment. E: Unique Environment

Junjian Yi and James Best () Gene and Environment 13 / 57 Univariate ACE Model The Univariate ACE Model: Basic Setup

Y , A, C, E are Normal and standardised therefore:

2 2 2 Var(Yi ) = 1 =a Var(Ai ) + c Var(Ci ) + e Var(Ei )

+2 [accov(Ai , Ci ) + aecov(Ai , Ei ) + cecov(Ci , Ei )]

corr(Ai , Ei ) and corr(Ai , Ei ) = 0 by deﬁnition. Another assumption:

Assumption 2: corr(Ai , Ci ) for all i Given Assumptions 1 and 2:

1 = a2 + c2 + e2 (2) and a2 is the variance in Y accounted for by genes. c2 is the variance in Y accounted for by shared environment. e2 is the variance in Y accounted for by non-shared environment.

Junjian Yi and James Best () Gene and Environment 14 / 57 Univariate ACE Model The Univariate ACE Model: Estimating a2, c2 and e2

corr(Y1, Y2) = cov(Y1, Y2) = cov((aA1 + cC1 + eE1), (aA2 + cC2 + eE2)) 2 2 2 = a cov(A1, A2) + c cov(C1, C2) + e cov(E1, E2)

+ ac [cov(A1, C2) + cov(C1, A2)]

+ ae [cov(A1, E2) + cov(E1, A2)]

+ ce [cov(C1, E2) + cov(E1, C2)]

Three more assumptions:

Assumption 3: corr(Ai , Cj ) = 0 for all i 6= j Assumption 4: corr(Ai , Ej ) = 0 for all i 6= j Assumption 5: corr(Ci , Ej ) = 0 for all i 6= j

Recall cov(E1, E2) = 0 by deﬁnition and we get:

2 2 corr(Y1, Y2) = a cov(A1, A2) + c cov(C1, C2) (3)

Junjian Yi and James Best () Gene and Environment 15 / 57 Univariate ACE Model The Univariate ACE Model: Twin Studies

Make assumptions (derived from Mendelian genetics) on the diﬀerence in genetic similarity between Monozygotic (identical) (MZ) and Dizygotic (non-identical) (DZ) twins to extract the loadings on the various factors. Also requires an assumption on twins shared environments: MZ DZ Twin Assumption 1: corrA1, A2) = 1; corr(A1, A2) = 0.5 MZ DZ Twin Assumption 2: corr(C1, C2) = corr(C1, C2) = 1 MZ twins MZ 2 2 corr(Y1, Y2) = a corr(A1, A2) + c corr(C1, C2) = a2 + c2

DZ twins DZ 2 2 corr(Y1, Y2) = a corr(A1, A2) + c corr(C1, C2) = 0.5 ∗ a2 + c2

Junjian Yi and James Best () Gene and Environment 16 / 57 Univariate ACE Model The Univariate ACE Model: Twin Studies

Subtracting correlations:

MZ DZ 2 corr(Y1, Y2) − corr(Y1, Y2) = 0.5a

Therefore Additive Genetics:

2 MZ DZ a = 2[corr(Y1, Y2) − corr(Y1, Y2) ]

Common Environment:

2 DZ MZ c = 2corr(Y1, Y2) − corr(Y1, Y2)

Unique Environment:

2 MZ e = 1 − corr(Y1, Y2)

Junjian Yi and James Best () Gene and Environment 17 / 57 Univariate ACE Model The Univariate ACE Model: Adoption Studies

Make assumptions on the diﬀerence in genetic similarity between Ordinary siblings (OD) and Adoptive siblings (AD)1. Also requires an assumption on siblings shared environments: OD AD Adoption Assumption 1: corr(A1, A2) = 0.5; corr(A1, A2) = 0 OD AD Adoption Assumption 2: corr(C1, C2) = corr(C1, C2) =1 Ordinary siblings OD 2 2 cov(Y1, Y2) = a corr(A1, A2) + c corr(C1, C2) = 0.5 ∗ a2 + c2

Adopted siblings AD 2 2 cov(Y1, Y2) = a corr(A1, A2) + c corr(C1, C2) = c2 1Adoptive siblings refers to siblings by adoption not siblings who have gone to diﬀerent families. For the other case the adopted siblings would be assumed to have no correlation in shared environment. Junjian Yi and James Best () Gene and Environment 18 / 57 Univariate ACE Model The Univariate ACE Model: Adoption Studies

This gives the following loadings on the diﬀerent factors: Additive Genetics

2 OD AD a = 2[corr(Y1, Y2) − corr(Y1, Y2) ]

Common Environment

2 AD c = corr(Y1, Y2)

Unique Environment

2 OD AD e = 1 − 2corr(Y1, Y2) + corr(Y1, Y2)

Junjian Yi and James Best () Gene and Environment 19 / 57 Univariate ACE Model Canonical Results of Behavioural Genetics (Sacerdote 2011) Canonical Results of Behavioural Genetics (Sacerdote 2011)

Junjian Yi and James Best () Gene and Environment 20 / 57 Univariate ACE Model Canonical Results of Behavioural Genetics (Sacerdote 2011) Canonical Results of Behavioural Genetics (Sacerdote 2011)

Junjian Yi and James Best () Gene and Environment 21 / 57 Univariate ACE Model Canonical Results of Behavioural Genetics (Sacerdote 2011) Canonical Results of Behavioural Genetics (Sacerdote 2011)

Junjian Yi and James Best () Gene and Environment 22 / 57 Univariate ACE Model Interpreting the results Interpreting the Results

1 Are the assumptions valid? 2 How representative are twins and adoptees of the population? Goldberger (1979) and Kamin and Goldberger (2002) criticise the representativeness of the samples. Holmlund, Lindahl and Plug ( JEL, 2011) find twin studies to be the most representative for the effect of parent education. 3 What are the policy implications? Goldberger (1979) and Manski (2011) claim that the variance analysis results cannot be relevant to policy. They are wrong. For example, the results imply that variance in parent wealth does not have a particularly large impact on child earnings. Adoption studies will not pick up the effects of environment prior to adoption on children: this could be large. The samples of adoptive families do not have a significant number of very low income families in the sample.

Junjian Yi and James Best () Gene and Environment 23 / 57 Univariate ACE Model Interpreting the results Examining the Assumptions

Assumption 1: Additivity GxE: Sensitivity to diﬀerent environmental factors rGE Epigenetics

Assumption 2: corr(Ai , Ci ) = 0 If raised by biological family this is unlikely: more so the more important genes are to behaviour. Non-Random Adoption Environment is aﬀected by one’s genes. Epigenetics

Junjian Yi and James Best () Gene and Environment 24 / 57 Univariate ACE Model Interpreting the results Examining the Assumptions

Assumption 3: corr(Ai , Cj ) for all i 6= j The phenotype or behaviour of one child can aﬀect how the parents treat both children.

Assumption 4: corr(Ai , Ej ) for all i 6= j The phenotype or behaviour of child i enters child j’s environment and child i’s behaviour is aﬀected by their genotype.

Assumption5: corr(Ci , Ej ) for all i 6= j The effect of the shared environment on i affects child i and enters into child j’s environment through child i’s behaviour. Something that happens to child j and not child i may affect the way the parents treat both of them.

Junjian Yi and James Best () Gene and Environment 25 / 57 Univariate ACE Model Interpreting the results Examining the Assumptions

MZ DZ Twin Assumption 1: corr(A1, A2) = 1; corr(A1, A2) = 0.5 Assortative Matching. Epigenetic considerations. MZ DZ Twin Assumption 2: corr(C1, C2) = corr(C1, C2) = 1 Diﬀerential Treatment of children.

Junjian Yi and James Best () Gene and Environment 26 / 57 Univariate ACE Model Interpreting the results Examining the Assumptions

OD AD Adoption Assumption 1: corr(A1, A2) = 0.5; corr(A1, A2) = 0 Non-random matching between adoptive parents. OD AD Adoption Assumption 2: corr(C1, C2) = corr(C1, C2) = 1 Diﬀerential Treatment of children.

Junjian Yi and James Best () Gene and Environment 27 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005)

Use a wider variety of sibling types which allows them to relax:

Assumption 3: corr(Ai , Cj ) for all i 6= j (Model II). Twin/Adoption Assumption 1: corr(A1, A2) for non-MZ biological siblings (Model III). Twin/Adoption Assumption 2: corr(C1, C2) = 1 for children in the same family (Model IV). They estimate four diﬀerent models using the following sibling types: 1 MZ twins reared together 2 MZ twins reared apart 3 DZ twins reared together 4 DZ twins reared apart 5 Non-twin full siblings reared together 6 Non-twin full siblings reared apart 7 Half siblings reared together 8 Half siblings reared apart 9 Adoptive siblings

Junjian Yi and James Best () Gene and Environment 28 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model I

Model 1: Standard model

corr(A1, A2) = 1 for MZ twins corr(A1, A2) = 0.5 for DZ twins and full siblings corr(A1, A2) = 0.25 for half siblings corr(A1, A2) = 0 for adoptive siblings corr(C1, C2) = 1 for all types of siblings reared together corr(C1, C2) = 0 for all types of siblings reared apart

Junjian Yi and James Best () Gene and Environment 29 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model I

Junjian Yi and James Best () Gene and Environment 30 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model I

Junjian Yi and James Best () Gene and Environment 31 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model II

Model II: Allow corr(Ai , Cj ) 6= 1. Those with genes conducive to high earnings also tend to have advantaged environment

corr(A1, C2) for biological siblings reared together, for siblings reared apart, and for adoptive siblings Find corr(A1, C2) to be insigniﬁcantly negative.

Junjian Yi and James Best () Gene and Environment 32 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model II

Junjian Yi and James Best () Gene and Environment 33 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model II

Junjian Yi and James Best () Gene and Environment 34 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model III

Model III: Allow assortative mating

corr(A1, A2) = 1 for MZ twins corr(C1, C2) = 1 for all types of siblings reared together corr(C1, C2) = 0 for all types of siblings reared apart Add three parameters to be estimated B corr(A1, A2) for DZ twins and full siblings: corr(A1, A2) = 0.43 and S corr(A1, A2) = 0.39. B corr(A1, A2) for half siblings: corr(A1, A2) = 0.25 and S corr(A1, A2) = 0.26. B corr(A1, A2) for adoptive siblings: corr(A1, A2) = 0.14 and S corr(A1, A2) = 0.18.

Junjian Yi and James Best () Gene and Environment 35 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model III

Junjian Yi and James Best () Gene and Environment 36 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model III

Junjian Yi and James Best () Gene and Environment 37 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model IV

Model IV: Relax the assumption that similarity of shared environment

corr(A1, A2) = 1 for MZ twins corr(A1, A2) = 0.5 for DZ twins and full siblings corr(A1, A2) = 0.25 for half siblings corr(A1, A2) = 0 for adoptive siblings corr(C1, C2) = 1 for MZ twins (normalized) Add three parameters to be estimated B corr(C1, C2) for DZ twins reared together: corr(C1, C2) = 0.406 and S corr(C1, C2) = 0.282. B corr(C1, C2) for non-twin reared together: corr(C1, C2) = 0.461 and S corr(C1, C2) = 0.34. B corr(C1, C2) for siblings reared apart: corr(C1, C2) = 0.209 and S corr(C1, C2) = 0.254.

Junjian Yi and James Best () Gene and Environment 38 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model IV

Junjian Yi and James Best () Gene and Environment 39 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Model IV

Junjian Yi and James Best () Gene and Environment 40 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Summary

Relaxing Assumption 3 or Twin/Adoption Assumptions 1 have little impact on estimates of heritability and shared environment eﬀects. Relaxing Twin/Adoption Assumptions 2 by allowing variance in corr(C1, C2) across sibling types decreases heritability estimate and increases shared environment estimates. A signiﬁcant role for genetic component of earnings variance: Smallest estimates: Males 20% (SE = 16%); females 10% (SE= 9%). Largest estimates: Males 28%; females 25% (SE = 8%). A role for shared environment component of earnings variance: Largest estimate: Males 16% (SE = 16%); females 18% (SE = 4%). Smallest estimate: Males 4% (SE = 4%); females 1% (SE = 4%). The importance of non-shared environment 64% of the earning variation is explained by neither genetic nor environmental resemblance

Junjian Yi and James Best () Gene and Environment 41 / 57 Univariate ACE Model Bjorklund, Jantti and Solon Bjorklund, Jantti, and Solon (2005): Questions and Criticisms

In Model II is is correct to interpret corr(A1, C2) 6= 1 as capturing the consideration that ‘those with genes conducive to high earnings also tend to have advantaged environments”? Why are the standard errors so large given the size of the samples used, particularly in Model IV? Model IV looks like it might suﬀer from strong multi-collinearity between the genetic and shared component. Model IV has estimates for the similarity in shared environments for diﬀerent sibling types that seem a little implausible.

Junjian Yi and James Best () Gene and Environment 42 / 57 Parent-Child Correlations: Adoption as a Natural Experiment Bjorklund, Lindahl and Plug (2006)

Use Swedish Adoption Data to estimate pre and post birth effects of family using adopted and biological child correlations with parent characteristics. They estimate: 1 Biological Child: bc bp bc Yi = β0 + β1Yi + vi 2 Adopted Child: ac bp ap ac Yj = α0 + α1Yj + α2Yj + vj Y is outcome of interest a = Adoptive b = biological c = child p = parent vi = unobserved child specific effect

Junjian Yi and James Best () Gene and Environment 43 / 57 Parent-Child Correlations: Adoption as a Natural Experiment

Junjian Yi and James Best () Gene and Environment 44 / 57 Parent-Child Correlations: Adoption as a Natural Experiment Bjorklund, Lindahl and Plug (2006)

There are several problems with this approach: 1 Non-Random Assignment: Correlation between biological and adoptive parent characteristics. Approximately 10% correlation between father earnings and 15% between parent education. 2 Delayed Adoption: There is little data on age of adoption. Late adoption will mean it is not a test of post and pre-birth eﬀects but of post and pre-adoption studies. 3 Comparable Samples: Pre-adoption environment is worse than average. Post-adoption environment is better than average. Possible Adoption Eﬀects

Junjian Yi and James Best () Gene and Environment 45 / 57 Parent-Child Correlations: Adoption as a Natural Experiment Sacerdote (2007)

Uses data on Holt Korean American Adoptees to control for non-random parent child matching. The Korean American Adoption system used a ﬁrst come-ﬁrst serve queuing system. Sacerdote provides evidence that the placement is random. Does not have data on adopted children’s biological parents. Estimates: 1 bc bp Yi = γ0 + βYi + γ1Malei + γ2Ai + γ3Ci + vi

2 ac ap Yj = γ0 + αYj + γ1Malej + γ2Aj + γ3Cj + vj

Aj = Full set of single year of age dummies

Cj = Full set of cohort dummies.

Junjian Yi and James Best () Gene and Environment 46 / 57 Parent-Child Correlations: Adoption as a Natural Experiment Sacerdote (2007)

Junjian Yi and James Best () Gene and Environment 47 / 57 Parent-Child Correlations: Adoption as a Natural Experiment How Representative is the Sample

There are the usual issues with how representative adoptees are of the general population. Parents have to go through an application process that takes 12-18 months (a long time to change their minds). Must ﬁle an application. Participate in a home study assessment. Adoption education classes. Pass a criminal background check. Parents between 25 and 45. Parents income is a minimum of 125 percent of the poverty threshold (data contains families near this threshold).

Junjian Yi and James Best () Gene and Environment 48 / 57 Parent-Child Correlations: Adoption as a Natural Experiment Review of Education Correlations: Sacerdote (2011)

Junjian Yi and James Best () Gene and Environment 49 / 57 Parent-Child Correlations: Adoption as a Natural Experiment Review of Earning Correlations: Sacerdote (2011)

Junjian Yi and James Best () Gene and Environment 50 / 57 Epigenetics What is Epigenetics

Some useful, quick, introductory literature: “Perceptions of Epigenetics” Adrian Bird (Nature, 2007) “Transgenerational Epigenetic Inheritance” Rakyan and Whitelaw (2006) “Environmental Epigenetics” Bollati and Baccarelli (Heredity, 2010) “Epigenetic germline inheritance” Chong and Whitelaw (2004) Several meanings but can be classiﬁed into two rough catagories (Bird, 2007): Conrad Waddington (1957): The process by which the genotype gives rise to the phentype. This is developmental biology (developmental biologists do not refer to their study as the study of epigenetics). Arthur Riggs: Mitotically and Meiotically heritable changes that cannot be explained through DNA.

Junjian Yi and James Best () Gene and Environment 51 / 57 Epigenetics Behavior Genetics and Post Genomics (Charney, 2011) Non-genetic Impacts on Phenotypic Variation

Retrotransposons (“Jumping Genes”): The movement of retrotransposons can be affected by environment. The movement of retrotransposons will be different in different individuals. Are MZ twins really identical? Some evidence that retrotransposon movement affects phenotype. Copy Number Variation Evidence that CNV can cause phenotypic changes, such as Autism, Schizophrenia and ADHD. Up to 10% variation between twin pairs in CNV. Aneuploidy Can cause disabilities and is not uniquely determined by DNA. Mitochondrial DNA Mitochondrial DNA can be the source of phenotypic variation. mDNA is non-Mendelian.

Junjian Yi and James Best () Gene and Environment 52 / 57 Epigenetics Behavior Genetics and Post Genomics (Charney, 2011) The Epigenome

The process of Methylation (and other processes) eﬀect which genes are expressed. The process of Methylation can be aﬀected by environmental factors. Methylation of a gene in one cell can be mitotically and even meiotically reproduced. For example the Perinatal environment of rats. If Dams lick and groom pups they have lower stress adult. There is evidence that this is caused by methylation. “Adopted” pups of Dams are more likely to lick and groom their own pups. There is some evidence that this is due to the transmission of methylated genes.

Junjian Yi and James Best () Gene and Environment 53 / 57 Epigenetics Behavior Genetics and Post Genomics (Charney, 2011) The Maternal Environment

The maternal environment begins when the egg is produced and its eﬀects begin when it is fertilised. The maternal environment will determine epigenetic transmission. MZ twins will be exposed to a more similar epigenetic environments than DZ twins and both are more similar than singletons. There is also variation within MZ twins and DZ twin dependent on the period of splitting.

Junjian Yi and James Best () Gene and Environment 54 / 57 Epigenetics Behavior Genetics and Post Genomics (Charney, 2011) Implications for Twin Studies

The correlation in the biologically determined features of twins and singletons are not determined by mendelian genetics alone. The correlation in biologically determined features can change over time. It is not accurate to split all determinants of outcomes into G, E and GxE. There is another factor that enters: the epigene.

Junjian Yi and James Best () Gene and Environment 55 / 57 Epigenetics Behavior Genetics and Post Genomics (Charney, 2011) Problems for Twin Studies: Defences and Responses

Defence by distinction: The A variable can capture the epigenetic effect as well as the genetic effect. However, this assumes that the epigenetic component does not affect corr(A1, A2) which is not necessarily reasonable. Underestimated Heritability Defence: If anything epigenetic considerations imply that the genetic effects are biased downwards. This is not clear as epigenetic effects can cause concordance or discordance and there are differing levels of epigenetic similarities between MZ and DZ twins. Similar Average Effects: If the epigenetic effects are similar for MZ and DZ twins the heritability estimates will not be biased. There are differing levels of epigenetic similarities between MZ and DZ twins.

Junjian Yi and James Best () Gene and Environment 56 / 57 Epigenetics Behavior Genetics and Post Genomics (Charney, 2011) My Thoughts on Epigenetics

There is evidence that epigenetics considerations have consequences for corr(A1, A2) if we interpret A as a measure of biometric variation rather than genetic variation. The pre, peri and post-natal environment can affect the expression of genes. This is similar to GxE effects and corr(G, E ) 6= 0 where G is taken to be a measure of biometric variation. The environmental effects on one person can be passed to their children: Grandchild twin studies? I worry about how large the effects of epigenetic variation are relative to genetic variation. Research in epigenetics seems to have found evidence for some effect but not evidence that it has a large impact relative to genetic heritability.

Junjian Yi and James Best () Gene and Environment 57 / 57