Improving estimation in genetic models using prior information

Espen Moen Eilertsen

doi:10.5324/nje.v26i1-2.2017

Abstract

Statistical models used to investigate research questions in behavioral genetics often require large amountsof data. This paper introduces some key concepts of Bayesian analysis and illustrates how these methodscan aid model estimation when the data does not provide enough information to reliably answer researchquestions. The use of informative prior distributions is discussed as a method of incorporating informationfrom other sources than the data at hand. The procedure is illustrated with an ACE model decomposition ofthe variance of antisocial personality disorder. The data originates from the Norwegian Twin Registry, andincludes adult twins assessed with the Structured Interview for DSM Personality (SIDP-IV). Inclusion ofprior information lead to a shift with respect to conclusions about the presence of shared environmentaleffects compared to a traditional analysis. Small and medium sized studies should consider use of priorinformation to aid estimation of population parameters.

Highlights

In behavioral genetics, it is often of interest to investigate the relative genetic and environmental contribution to phenotypic variance
Because it is known that monozygotic (MZ) twins share all their genes whereas dizygotic (DZ) twins on average share half of their genes, genetic contributions to a phenotype can be investigated by assuming equal environmental influences in MZ twins and in DZ twins (Plomin et al, 2013)
Additive genetic effects seem to account for roughly half of the variation in antisocial personality disorder (ASPD) for males, whereas there is not much evidence suggesting the presence of shared environmental effects

Summary

INTRODUCTION

It is often of interest to investigate the relative genetic and environmental contribution to phenotypic variance. Additive genetic effects seem to account for roughly half of the variation in ASPD for males, whereas there is not much evidence suggesting the presence of shared environmental effects Either, this is because shared environments do not contribute to variation in the population investigated, or, the data is too noisy to accurately estimate a variance component attributable to shared environmental effects. In order to form informative prior distributions for the variance parameters, results from Rhee and Waldman’s (2002) meta-analysis on the genetic and environmental influences on antisocial behavior were used. This did not lead to any substantial change in parameter estimates.) Based on this decision, gamma distributions with mode equal to the meta-analytic aggregate results and standard deviations equal to .15 were used as prior distributions for the variance components (see figure 2). In many ways this seems like a less drastic decision, because it seems unreasonable to conclude that there exist no common environmental effects in the population

DISCUSSION

COMPLIANCE WITH ETHICAL STANDARDS