Online Program

Balanced Response and Nonresponse Weighting Adjustments
*Carl-Erik Särndal, Statistics Sweden and Örebro University 


Many surveys today are affected by high nonresponse, viewed by all parties concerned as an impediment to survey quality, in business surveys as in surveys on individuals and households. Faced with the problem, statisticians need to intervene (a) at the data collection (DC) stage, and (b) at the estimation (EST) stage. The purpose of the lecture is to examine these tasks in the light of the available extraneous information, the auxiliary variables (AV). Unlike the study variables affected by nonresponse, the AV are known for the responding units in the selected probability sample as for the non-responding ones, and therefore crucially important. The objectives for (a) and (b) interact. The DC is monitored, in the spirit responsive design, by indicators of balance and representativity (of the set of respondents), and distance (between respondents and non-respondents), computable on selected AV. Once the DC terminated, the EST stage needs to focus on reducing as far as possible the bias that still affects the estimates, despite any balancing at the DC stage. Adjustment by weighting uses calibration on a suitable set of AV, including possibly those having served already in balancing.