Analyzing Methods for Aggregate-Disaggregate Data Fusion
This thesis examines methods for aggregate-disaggregate data fusion. In the presence of both accurate, aggregate-level data which often lacks granularity, and disaggregate but potentially biased individual-level data, marketers and statisticians alike struggle in determining the appropriate weight to assign to each piece of information. Through simulation, this paper tests the effectiveness of the use of a multivariate normal approximation to aggregate-level data. An alternative algorithm to impute the missing data, subject to aggregate characteristics, and then estimate parameters is posed and evaluated against the initial method. The paper finds that the multivariate normal approximation not only outperforms the proposed algorithm but also quite accurately estimates model parameters.