Using Community-Level Correlates to Evaluate Nonresponse Effects in a Telephone Survey