Learning which demographic attributes (race, age, etc.) are most influential in predicting how annotators will judge subjective content.