Assessing the effects of cooperation bias and attrition in behavioral genetic research using data-weighting

Because twins and adoptees are a rare resource, they are often studied repeatedly over a period of many years. Differential attrition, and in some studies initial cooperation bias, have the potential to lead to serious biases to estimates of genetic and environmental parameters. Since non-response is often influenced by multiple binary or categorical sociodemographic variables, maximum-likelihood methods are not easily adapted to adjust for such effects. In this brief note we illustrate the use of data-weighting to assess the likely effects of cooperation bias or attrition both on measures of mean or prevalence, and on twin pair correlations or concordances, using data from the Australian twin panel 1981 survey and alcohol challenge studies. Participants in the alcohol challenge study were on average younger, more socially nonconforming, heavier drinkers, more likely to be unmarried, and less likely to report their religion as Other Protestant. Reweighting the alcohol challenge sample to have the same distribution on these variables as the Australian twin panel 1981 survey respondents confirmed that individuals who would feel very intoxicated after a challenge dose of alcohol were underrepresented in the study. However, pairwise data-weighting indicated that this cooperation bias was leading to only a slight underestimation of the importance of genetic effects on subjective intoxication.