Date of this Version
Variables used in nonresponse adjustments are sometimes associated with the response outcome and sometimes with survey variables. Both associations are paramount to the reduction of nonresponse bias in survey estimates. We demonstrate that it is only when both are present that adjustments change the point estimate of the mean from the unadjusted mean. We also set out to test whether the relatively low associations commonly found in survey data are sufficient to achieve this goal. There are many such auxiliary variables that can be used for nonresponse adjustment. We demonstrate augmenting nonresponse adjustments in the 2004 National Election Study using interviewer observations on the sampled address and the contacted individual.