The context paradox in attitude surveys: assimilation or contrast?