Data fusion poses challenging methodological issues for inferring the joint distribution of two random variables when the information available is mainly confined to the marginal distributions. When the variables are categorical, the challenges are even more severe. Applications of categorical data fusion are of top importance in marketing, especially in advertising. A great deal of categorical data fusion methods are confined to binary variables. In this paper we develop an innovative approach to categorical data fusion that extends previous methodologies and applies to categorical variables with any number of levels. We introduce a new concept of "evident dependence" that describes a variety of patterns of joint distributions given the marginals. Using information from partially fused data, our method smoothly accommodates a Bayesian approach based on mixtures of joint distributions constructed using evident dependence. The approach is illustrated using data from the advertising industry.
- Evident dependence
- Mixture modeling
ASJC Scopus subject areas
- Economics, Econometrics and Finance (miscellaneous)