A SUBSAMPLING APPROACH TO NON-RESPONSE FOLLOW UP
The ABS recognises a pressing need to reduce overall survey costs. Field costs are a very large component of overall survey costs and are driven up by the requirement to deliver high response rates. The effort required to reach response rate targets has been increasing over time. Current ABS research aims to provide a strategy for cost reduction in data collection that manages the impacts on estimate quality. In this context we consider quality in terms of sampling error (variance) and non-response error (non-response bias).
Structural cost savings can be realised, inter alia, with a focus on the organisation of intensive follow up (IFU) efforts. The specific method under consideration is sub-sampling for IFU. After initial contact and a request to complete a survey are made to each unit in the selected sample, any non-responding units are currently followed up in an effort to elicit response. The idea behind sub-sampling for IFU is to instead focus the IFU effort (or a fraction of the existing IFU effort) to only a fraction of the non-responding units.
The sub-sampling approach is supported by estimation methods that make use of the distinction between units that respond to the initial request and those that respond through follow-up, or, for categorical variables, the constancy in the ratio of propensities between the initial and IFU respondents. We have the choice of some units receiving IFU with certainty (e.g. influential businesses, or units first joining a panel) while the selection probability of other units can be optimised with respect to criteria of interest (e.g. creating a sample with more even response propensities).
For more information, contact Bruce Fraser Methodology@abs.gov.au