Ascertainment of the Twin Sample

There is now strong evidence that general intelligence as measured by ability tests is largely genetically determined, in western white societies at least (Eaves, 1975). The evidence for educational achievement is more conflicting and several studies have found lower heritabilities in school tests and examinations. In this study we analyse the results of a sample of monozygotic (MZ) and dizygotic (DZ) brins who sat for the Intermediate examination of the Public Examinations Board of South Australia. This examination was taken by about two-thirds of the 15-year-old cohort in South Australia until 1968, after which it was abolished. A candidate was generally expected to have spent three years at secondary school preparing for examination in about seven subjects. The twin method has been subjected to close scrutiny and considerable criticism which has not abated even in the centenary year of its inception by Galton in 1875 (Schwartz & Schwartz, 1974). Many of these criticisms have been dealt with at length (Vandenberg, 1966; Mittler, 1971; Eaves & Eysenck, 1975). Nevertheless, before accepting results of analyses of twin data it is . important to try to estimate the effects of any biases in the ascertainment and composition of the twin sample. One aim of biometrical genetics is to obtain estimates of parameters which will adequately describe observed phenotypic variation in a given character. In the absence of genotypeenvironment interactions, Jinks & Fulker (1970) show that initially the total variance may be partitioned into G1 = within-family genetic component, G2 = between-family genetic component, E1 = within-family environmental component, E2 = between-family environmental component. In the case of twin studies, we can see that biased ascertainment will generally restrict the sampling range and cause underestimation of the between families components E2 and G2 , while incorrect zygosity diagnosis could alter the values of E1 and G1> the "rithin families components, and consequently affect the between families components as well. In this paper we attempt to estimate the importance of these two kinds of bias in this study.