The Use of GIS Information as Auxiliary Data for Nonresponse Bias Analysis
Abstract:
The Nielsen Company, like most other market research firms, is concerned with falling response rates and the threat of nonresponse bias. The challenge in understanding and adjusting for nonresponse is always the lack of information about the non-responding cases. In recent years, researchers have considered the use of paradata, interviewer observations, and aggregate Census information to provide information about the non-responders. This study introduces and evaluates the use of a new type of GIS-based data, called POI (Points-of-Interest), as a potential auxiliary data source for nonresponse bias analysis. The appeal of auxiliary data is that it is available at the frame level (i.e. for responders and non-responders) and therefore it can be potentially used for computing post-survey nonresponse adjustments. To evaluate nonresponse bias in its week-long television viewing diary survey, Nielsen conducted a nonresponse bias survey in 2012. Results revealed significant differences in demographic characteristics and TV viewing measures between diary responders and non-responders. After geocoding the address of each selected case (respondents and nonrespondents), we merged in information about the distance of each household to local points of interest such as hospital, grocery store, airport, etc. From multivariate analyses, POI variables were found to be less effective in reducing nonresponse bias, and their effectiveness varied as a function of the TV viewing measure and geography. These findings will likely have important implications for the usage of POI data in computing post-survey nonresponse adjustments for mitigating nonresponse bias.
Recommended Citation:
Rao, K., & Eckman, S. (2013). The Use of GIS Information as Auxiliary Data for Nonresponse Bias Analysis. Paper presented at the Midwest Association for Public Opinion Research, Chicago, IL.
Attached Documents:
- MAPOR 2013 Program (see page #12 for the mention)
- For a copy of this presentation, please send me a comment with your email address in the box below.
Would be interested to read the paper. I am currently working on a project which is attempting to use POI data to explore nonresponse bais in the Euroopean Social Survey in the UK. We are currently at an early stage – selecting which POIs to use in our models – and any insights from other studies would be much appreciated