Field Substitution and Unit Nonresponse
Field substitution is used to compensate for unit nonresponse in sample surveys. Unlike
methods such as weighting or imputation, substitution preserves the designed (optimal)
structure of the sample, and this can be a source of certain advantages over the
alternative methods. The most important advantage is the potential gain in precision,
particularly when two-stage cluster samples are used. However, due to specific properties
of the increase in sampling variance, this gain is usually small, and practitioners who
use substitution often overlook this fact. On the other hand, substitution introduces an
extra bias into the estimates. Additionally, severe practical difficulties arise in the
field work process, and these will generally outweigh the benefits of this approach. Thus,
despite its relatively frequent use in probability samples, the substitution can rarely be
justified. Nevertheless, specific circumstances in which this practice can be
theoretically and practically advantageous do exist. We encounter such situations in
samples with small take per cluster or stratum, where efficient field controls are
provided, as with telephone surveys.
Missing data; survey methodology.