Tetrachoric correlations, as well as estimation of heritability and best-fitting etiological model by means of structural equation modeling, were carried out using computer software (MX; Michael C. Neale, Department of Psychiatry, Virginia Commonwealth University; available at www.vcu.edu/mx/).
13 These calculations are based on a liability model, which assumes that the dichotomous distribution of cilioretinal arteries (present versus not present) reflects an underlying normally distributed liability of the population. When a threshold value of the liability is exceeded, an individual is affected, otherwise not. The threshold reflects the prevalence of the trait. These are standard assumptions in quantitative genetic analysis of categorical traits.
14 The classical twin model is based on the assumption that MZ twin pairs share all their genes, while DZ twin pairs, like other siblings, share on average 50% of their genes, and that both MZ and DZ twin pairs share a common environment to the same extent. This means that a greater similarity among MZ than among DZ twin partners for a certain phenotype can be due only to genetic factors.
15 Structural equation modeling quantifies sources of individual variation by decomposing the observed phenotypic variance into genetic and environmental variance. The genetic contribution can be further divided into an additive (
A) genetic variance component (representing the influence of alleles at multiple loci acting in an additive manner) and a nonadditive (
D) genetic variance component (representing intralocus interaction [dominance] and interlocus interaction [epistatis] of alleles). The environmental component is subdivided into a common (
C) environmental variance component (representing environmental factors affecting both twins in a pair, a source of similarity) and a nonshared or random (
E) environmental variance component (representing environmental factors not shared by twins, a source of dissimilarity). The latter component (
E) also includes random factors and measurement errors.
16 Heritability is defined as the proportion of the total phenotypic variance attributable to genetic variance. The components
C and
D cannot be estimated simultaneously in a twin study. We fitted the
ACE and
ADE models and the submodels
AE,
CE, and
E to the data. The criteria for best-fitting model were based on Akaike’s information criterion (AIC), goodness-of-fit χ
2 test, degrees of freedom, and
P-value. The model with the lowest negative AIC reflects the best balance between goodness of fit and parsimony.