CRM Colloquium

Colloquium of the Centre for Research in Mathematics

TITLE: 'Association and aggregation for contingency table analysis'

ABSTRACT: For reasons of confidentiality, government departments and other agencies often release information to the public that has been aggregated. When working with aggregated data it is important to ensure that one keeps in mind any loss of information (due to the aggregation process) when making conclusions at the individual level. There are a variety of techniques that exist which analyse aggregate data and these generally lie within the family of ecological inference techniques. This presentation is not about those techniques. Rather, we shall consider a very simple illustrative example to highlight the loss of information due to aggregation and show how correspondence analysis can be used as a means of visually identifying the source of such loss. This presentation will explore the impact of aggregation using the standard statistical techniques of simple linear regression and the chi-squared test of independence.