Background. Previous twin studies of hyperactivity have supported a ‘contrast effect’, whereby the more hyperactive the rating of one twin, the less the rating of the other. It has not been clear whether contrast effects occur in the twins' behaviour or in the ratings made of their behaviour but the implications for hyperactivity are different under the two models.

Method. We use hyperactivity ratings from mothers and teachers for 1644 twin pairs in the Virginia Twin Study of Adolescent Behavioral Development (VTSABD) to explore the origin of contrast effects, making use of independent teacher reports in a proportion of twins. Models were fitted separately for the two informants and jointly to ratings combined through a latent variable, using structural equation modelling.

Results. Models for maternal data confirm the contrast effect previously reported. Teacher ratings show a different form of bias, with both twin confusion and correlated errors representing alternative but not mutually exclusive explanations of the data. Latent variable modelling of the joint responses allowed comparison of a model in which the contrast effect was placed on maternal ratings, representing bias, versus one in which the contrast occurred in the underlying ‘true’ phenotype. The fit of the former model was significantly better.

Conclusions. Support is provided for the notion of contrast effects as a form of rater bias in maternal hyperactivity ratings. Different bias in teacher reports highlight that no one report can be considered a gold standard. The extent to which such biases may distort information for other data sources such as sib-pair studies of concordance/discordance is discussed.