Exploring the effect of replacement levels on data fusion methods: A Monte Carlo simulation approach

Data fusion is a technique used for creating an integrated database by combining two or more databases that include a different set of variables or attributes. This paper attempts to apply data fusion technique to customer relationships management (CRM), in that we can not only plan a database structure but also collect and manage customer data in a more efficient way. In particular, our study is useful when no single database is complete, i.e., each and every subject in the pre-integrated database contains somewhat missing observations. According to the way of treating the common variables, donors can be differently selected for the substitution of the missing attributes of recipients. One way is to find the donor that has the highest correlation coefficient with the recipient by treating common variables metrically. The other is based on the closest distance by the correspondence analysis in case of treating common variables nominally. The predictability of data fusion for CRM can be evaluated by measuring the correlation of the original database and the substituted one. A Monte Carlo Simulation analysis is used to examine the stability of the two substitution methods in building an integrated database.