|
Abstract:
|
The main tool used in marketing mix models is regression analysis. In this specific environment "over-studied" in general regression obtains several features, which are not typically considered in statistical literature in such a combination. These are direct character of influence of some variables to outcome and indirect of others; importance of time decaying effects; high correlation between factors, which should be used together because of economical requirements; desirable maximization of number of factors in a model, as opposite to traditional reductionism; a very big number of possible factors combined with small number of available data points; as a result of previous points, serious difficulties in optimal variables selection. Paper provides some recommendations on these issues with results of Monte-Carlo simulation about validity of regression coefficients.
|
