Abstract:
|
Although commonly used in many surveys as a strategy to deal with unit nonresponse, substitution is frequently criticized and has received very little attention by survey researchers. In fact, there is evidence to suggest that the performance of substitution as a strategy to mitigate nonresponse is comparable to other adjustment methods, such as weighting or imputation (Vehovar, 1999; Rubin and Zanutto, 2002). However, as with many other nonresponse adjustment methods, research on and applications of this method has been limited to ignorable nonresponse mechanisms. This paper presents a substitution procedure that incorporates a nonignorable nonresponse mechanism in the selection of the substitutes of the nonresponding units through the use of pattern-mixture models. This method can be employed to perform sensitivity analysis of a range of missingness models using additional real data of the substitutes, as opposed to other methods that use predicted values under a model or data from hot-deck donors already present in the responding sample. This methodology is evaluated and compared to other nonresponse adjustment methods through a simulation study.
|