This paper by Mike Bankier of Statistics Canada explores the issues of editing and imputation (E&I) in the context of population census data. E&I should eliminate non-response and inconsistent responses in an effective but conservative manner. It should be possible to specify edits easily and then either do minimum change donor imputation or specify deterministic imputation actions. In addition, the user should be able to evaluate the results easily during Census production to determine if the imputation should be redone. Because usually not all census variables can be processed simultaneously, the variables should be partitioned into groups and then these groups should be processed sequentially or in parallel.
Somewhat different imputation techniques are often required for households with total non-response compared to those with partial non-response. Also, sampled households that received a long questionnaire with additional questions should be processed carefully to avoid biasing estimates based on these samples.
The above issues will be illustrated by describing the techniques used in the 2006
Canadian Census of Population to carry out E&I.
Paper presented at the Satellite Meeting on Innovative Methodologies for Censuses in the New Millennium of the 56th Session of the ISI, Lisboa, 31 August - 2 September 2007.
Optionally provide your comments to help us improve this article...
Thank you for your feedback!