Patient- or Physician-Oriented Marketing: What Drives Primary Demand for Prescription Drugs?

Recall that some of the unknown parameters in our brand sales model are specified as categoryand brand-specific while others are not (see Equation 6 in the paper). To estimate this model we have data available for 2,831 brands over a period of 21 quarters starting in quarter 4/2000 until quarter 4/2005. Since the brand sales model includes lagged sales the first quarter is inherently lost. In addition, we note that many brands are launched later than quarter 4/2000 and a few brands are withdrawn from the market during the observation period. As a result, we have an unbalanced panel data set amounting to eventually 47,308 observations which we can use for estimation. Hence, we have on average 16.7 quarters per brand and 32.9 brands per category (86 categories) available. Our brand sales model incorporates brand-specific effects which reflect brand heterogeneity in the market. Considering only own and competitive marketing expenditure variables, a fixed-effects specification would require estimating ~8,500 own brand sales effects and ~25,000 cross-effects associated with detailing, professional journal advertising, and DTC. With 47,308 observations at hand, we need to impose some structure on the heterogeneous