For causal inference with panel data, matrix completion with row-wise error bounds enables accurate per-unit treatment effect estimation without knowing treatment propensities—a significant improvement over methods that only guarantee average treatment effect accuracy.
This paper solves the problem of estimating how treatments affect individual units (rather than just average effects) using panel data. The researchers frame this as a matrix completion problem and develop a new method that achieves better error bounds than existing approaches, with theoretical guarantees that work even when treatment assignments are unknown and non-uniform.