A New Method for Partial Correction of Residual Confounding in Time-Series and Other Observational Studies