Adjustment for nonresponse bias in household surveys

Fannie Cobben PhD

Project at: Statistics Netherlands (CBS), Voorburg, The Netherlands

Supervisors: Prof. dr J.G. Bethlehem, prof. dr ir J.G. De Gooijer

Project running from: 1 March 2004 – 1 March 2009

Summary
At Statistics Netherlands, surveys are performed in order to produce statistics about the popula-tion. A random sample is selected from the population. When for some reason individuals that are selected in the sample do not provide the requested information, there is nonresponse. Non-response in surveys affects estimates of population characteristics if respondents and nonrespon-dents on average provide different answers to the survey questions. In order to detect and adjust potential bias in the estimates, auxiliary information is linked to the surveys for both respondents and nonrespondents. The method that is most commonly used at Statistics Netherlands to adjust for the nonresponse bias is linear weighting.
A vital part of the adjustment approach is the availability of auxiliary information. At Statistics Netherlands, more and more information becomes available from registers. This increasing amount of auxiliary information provides a lot of possibilities but at the same time it complicates selecting the most informative auxiliary variables and models.
This project considers alternative adjustment techniques for the nonresponse bias, like the propensity score method and sample selection models, as well as datamining techniques to select the most informative auxiliary variables.

Date of defence: 5 October 2009

Title of thesis: Nonresponse in sample surveys, methods for analysis and adjustment