INFO 188 Lecture Notes - Lecture 25: Jargon, Multiple Comparisons Problem, Implicit Stereotype
Document Summary
A bunch of ways we talk about the terms, but all valid in different ways. Would have to explicitly put in an extra keyword green to get a green banana. Compas did as well as randomly picking people to predict crimes. The compas data model was only really using gender, age, and priors. Problem with compas: solving a problem that is unsolvable. Criminality is not something you can observe. Just because you are a criminal doesn"t mean you are arrested, and just because you are arrested doesn"t mean you are a criminal. When you try to measure a concept, make sure you can actually measure that concept. Only needs historical crime data, type, location, and time to calculate background crime and aftershock crimes. Certain groups overrepresented as perpetrators and underrepresented as victims. Total crime = observed and unobserved crime >= observed crime >= reported and observed crime.