Despite advances made in broadening access to higher education, some evidence suggests that college completion rates have actually declined over time, and attainment gaps stemming from race/ethnicity and socioeconomic background have remained stubbornly persistent (Bound, Lovenheim, & Turner, 2010; National Center for Education Statistics, 2019). Colleges and universities are searching for ways to better identify students with lower odds of postsecondary success and develop programs to support their persistence and completion. This paper presents collaborative research from four different states -- Texas, New York, Virginia and Illinois -- examining the feasibility and accuracy of models that predict students' likelihood of degree completion using only administrative data on students' pre-college characteristics and early college experiences. The authors employed fairly simple statistical methods on statewide longitudinal data systems (also known as student unit records systems or "SURDS") to predict degree completion for undergraduate students. Their focus was on the overall predictive validity of these models, which is often neglected in educational research that focuses on relationships between student characteristics and degree completion (American Diploma Project, 2004; Adelman, 1997, 1999, 2006; Attewell, Heil & Reisel, 2011; Braxton, 2000; Chen & Carroll, 2005; Complete College America, 2011; Horn, Kojaku & Carroll, 2001; Tinto, 1994, 2012). [The paper was written with Christopher Maggio, Frederick Tucker, Jay Brooks, Xiaodan Hu, Tod Massa, Feng Raoking, and Nathan Wilson.]