Variation in response functions : a source of measurement error in attitude research