In embodiments of the present invention, a method is described for reducing bias by data fusion of a household panel data and a loyalty card data. In embodiments, a method is provided for receiving a consumer panel dataset in a data fusion facility, receiving a consumer point-of-sale dataset in a data fusion facility, receiving a dimension dataset in a data fusion facility, fusing the datasets received in the data fusion facility into a new panel dataset based at least in part on an encryption key, estimating a consumer behavior using a first model based on the consumer panel dataset, estimating a consumer behavior using a second model based only on those consumers present in both the consumer panel dataset and the consumer point-of-sale dataset, and refining the first model based at least on the results of the second model.