Ascertainment of a twin sample by computerized record matching, with assessment of possible sampling biases

We review progress made in ascertaining a twin sample by computerized record matching. A tar get sample of 1856 female like-sex twin pairs born 1975-1983, and believed to be both still living, was identified from state birth records. In 30 months, during 1995-1997, contact with 86% of families was established, although the success rate was lower for minority (principally African-American) families (74%). An estimated 15% of school-aged twin pairs were discordant either for school grade level or for school attendance and would have been missed in any school-based ascertainment scheme. Equal proportions of majority and minority families were living in state at the time they were traced (85.6 versus 85.3%). Using a conservative adjustment for the reduced probability of finding out-of-state ("mobile") families, we project that 80% of families were still living in the state where the twins were born. Mobile families on average had better-educated parents, had higher incomes, and were more likely to remain two-parent families but did not differ in rates of parental psychopathology (alcoholism or depression). Hard-to-find families-those in which time to completion of a first-contact interview was greater than the 90th percentile-were more likely to be African-American, to have an alcoholic biologic father, to be living out of state, and to be larger and poorer. Adjusting for undersampling of out-of-state residents had little impact on conclusions about genetic and environmental contributions to offspring risk of behavioral problems, as assessed by parental ratings of the twins' oppositional symptoms and school grades.