Converting Nonrespondents in PIAAC Germany 2012 Using Responsive Measures

Abstract

Using paradata to modify design features during fieldwork is the earmark of responsive designs (Groves & Heeringa, 2006). One objective of responsive approaches is to improve the composition of the final sample by gaining the participation of nonrespondents. A simple but innovative attempt at realizing such a response intervention was undertaken during the fieldwork of PIAAC Germany 2012. Different groups of nonrespondents were identified for follow-up efforts. With a view to the outcome measures of PIAAC, basic skills of the adult population, two groups were focused: Non-nationals and sample persons with low educational attainment. To identify these groups, different sources of auxiliary data were used (sampling frame, interviewer observations, and a commercial vendor database). Non-nationals were identified using information from the sampling frame. The challenge was to identify sample persons with (presumably) low levels of education. This was achieved by selecting a set of auxiliary variables, and subsequently using classification trees to model and predict sample persons with low levels of education. The sample persons were sent carefully crafted tailored letters during the re-issue phase. Overall, the cost-benefit balance of this intervention is rather disproportionate: A high level of effort with little apparent impact on the final sample composition. Nevertheless, this explorative endeavour was worthwhile and informative. In particular, the model-based prediction of different types of sample persons can be regarded as a promising approach.