Characterizing Fairness Over the Set of Good Models Under Selective Labels

             Amanda Coston * 1 Ashesh Rambachan * 2 Alexandra Chouldechova 3

             Abstract                ous methods for learning anew the best performing model
   Algorithmic risk assessments are used to inform      among those that satisfy a chosen notion of predictive fair-
   decisions in a wide variety of high-stakes settings.    ness (e.g. Zemel et al. (2013), Agarwal et al. (2018), Agar-
   Often multipl ...