This paper is about the fusion of a television people meter panel and a magazine readership/product usage survey in Mexico. During the process, two different methods of data fusion — unconstrained and constrained statistical matching — were evaluated on a number of criteria: computational complexity, donation frequency, sample sizes, matching success rates, preservation of audience and incidence levels, together with a number of individual-level and aggregate-level diagnostic tools. Collectively, these results provide a detailed look into the inner workings of data fusion.
We do not advocate the inherent superiority of either constrained or unconstrained matching. So our discussion will focus on how to make decisions based upon the trade-offs among various factors in a given situation, and this will therefore be of interest to researchers everywhere.
Symposium: 2001: Venice, Session 3 - Quality and Impact of Fusion
Authors: De Montigny, Michelle, Soong, Roland
Organisations: KMR Group
Topics: Fusion and Modelling