Christen, Markus, Sachin Gupta, John C. Porter, Richard Staelin and Dick R. Wittink (1997). "Using Market-Level Data to Understand Promotion Effects in a Nonlinear Model", Journal of Marketing Research, Aug 1997, 322-334

The authors show analytically, empirically, and numerically through simulation that the estimated effects from linearly aggregated market-level data differ substantially from comparable effects that are obtained from store-level data. The magnitude of this difference renders market-level data largely unsuitable for econometric modeling, unless the marketing manager compensates for the bias that results from the incompatible aggregation. The authors introduce a new approach, a relatively simple debiasing procedure derived from simulated data. They show that this debiasing approach results in substantially improved parameter estimates. They illustrate the value of the procedure by applying it to scanner data for powdered detergents and comparing the debiased parameter estimates to results obtained from store-level data and an alternative aggregation method that maintains homogeneity for selected promotional activities.

By only aggregating homogeneous entities, Marketing Analytics’ Store Group model eliminates "aggregation bias", and with it, huge potential coefficient errors. In this independent published study, promotional lift estimate error rates for market level models were 655% while the StoreGroup model error was 16%.