Efficient Low-Rank Estimation From Sparse Social Science Data