Information theory based measures of association applied to opinion polls


  1. Eles, P.T.
  2. Vincent, E.
Corporate Authors
Defence R&D Canada - Centre for Operational Research and Analysis, Ottawa ON (CAN)
A methodology is described for analyzing opinion poll data and, in particular, for measuring associations between poll questions. The methodology is based on information theory and relies on calculating the information content of each question and the information shared between questions. In this context, shared information is defined as the amount of information revealed about an individual’s response to one poll question once his/her response to another question is known. Information diagrams are introduced to facilitate interpretation of results. The paper focuses on defining a process for identifying the most significant associations between questions from polls that have a large number of questions. The key results of this work are: the definition of information theory based measures of association, the introduction of a step-by-step process for identifying relevant associations from large datasets, insights into the interpretation of calculated measures of association, a summary of lessons learned from practical application of the technique, and a demonstration of the technique with a real-world example taken from an opinion poll of the Afghan population in Kandahar Province commissioned by the Canadian Forces in February 2008.

Il y a un résumé en français ici.

Report Number
DRDC-CORA-TM-2008-022 — Technical Memorandum
Date of publication
01 Jul 2008
Number of Pages
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: