BackgroundAttrition in longitudinal studies often leads to healthier, more educated and less demographically representative samples at follow-up. The extent to which attrition impacts results from twin studies, and the efficacy of methods to correct for attrition, remains unclear. MethodsWe address these questions using data from the Twins Early Development Study (TEDS). First, we found that baseline variables, including twin sex, zygosity, ethnicity and maternal socioeconomic status, predicted 5–7.5% of the variance in follow-up participation. Next, we examined the impact of attrition by comparing estimated means, variances, correlations, and genetic and environmental influences before and after removing participants who did not participate at follow-up. We evaluated a range of twin measures from the Year 2, 12 and 16 time points, including internalising, externalising, educational and cognitive measures. ResultsAttrition affected means and variances, but had little material effect on correlations and genetic and environmental variance components. Finally, we evaluated the efficacy of two popular correction methods—inverse probability weighting and multiple imputation—to reduce bias, focusing on adolescent outcomes. Both methods partially reduced bias in mean estimates; however, imputation also introduced bias for some effects. ConclusionsOur results suggest that for the variables we examined in TEDS, correlations and genetic and environmental variance components are not strongly impacted by attrition. Imputation and weighting methods can reduce attrition bias when estimating means and variances but may introduce bias into estimates of association.