Current uses of value-added modeling largely ignore or assume away the potential for teachers to be more effective with one type of student than another or in one subject than another. This paper explores the stability of value-added measures across different subgroups and subjects using administrative data from a large urban school district. For elementary school teachers, effectiveness measures are highly stable across subgroups, with correlations upwards of 0.9. The estimated cross-subject correlation between math and English language arts is around 0.7, suggesting some differential effectiveness by subject. To understand the magnitude of this correlation, I simulate targeted re-sorting of teachers to classrooms based on their comparative advantage. The results suggest that using multiple measures of value-added to specialize teachers by subject could produce small average increases in student achievement, and larger increases for some students.