When Seeing Is Believing: Generalizability and Decision Studies for Observational Data in Evaluation and Research on Teaching