Statistics Seminars: Statistical disclosure control and log-linear modelling
12 February 2008 14:15 in CM221
Researchers need access to survey microdata files for analysis. However, in making such files available, the data collection agency needs to protect the confidentiality of the respondents. Even after deleting direct identifiers, such as name and address, it may still be possible for a user of a file to identify a respondent by matching a record to an external database using 'identifying' variables, such as age and occupation. This paper provides a brief introduction to 'statistical disclosure control' and discusses how the 'risk of identification' may be assessed when the identifying variables are categorical and the sample membership is unknown. The approach involves selecting a log-linear model for the identifying variables.
Contact firstname.lastname@example.org for more information