Dixon et al.: An information entropy approach to salience for survey-driven simulation

David S. Dixon, Pallab Mozumder, William F. Vásquez

Interpreting a 2010 survey of Houston-area households subject to hurricane Ike in 2008, the difficulty in inferring ex-ante preferences from ex-post survey responses is exacerbated by many of the responses that rated most or all of the issues as “extremely important”. From an information theoretical point of view, ranking all issues the same provides no more information than providing no answer at all. A response ranking only a few issues at a variety of levels of importance, on the other hand, provides insight about the individual respondent. In looking at the responses to ten related questions, information entropy – an information theoretical measure of information content – provides a measure of this variety in respondents. Assuming that higher information entropy implies higher salience, the group of respondents with highest information entropy are shown to be broadly representative of the total population surveyed while also providing clearer insights into motivations. A NetLogo agent-based model is calibrated based on the high information entropy survey responses.

Download PDF and Join our Discussion!