In this recipe, we use the movie dataset to identify the three observational units (movies, actors, and directors) and create separate tables for each. One of the keys to this recipe is understanding that the actor and director Facebook likes are independent of the movie. Each actor and director is mapped to a single value representing their number of Facebook likes. Due to this independence, we can separate the data for the movies, directors, and actors into their own tables. Database folks call this process normalization, which increases data integrity and reduces redundancy.