I am first trying to build a model using regression analysis to explain weekly sales, stripped of all exogenous variables (pricing, seasonality, etc) and all marketing expenditures . Currently, I have 3 independent variables that give an R-sq of about 85-90%. My next step is to add in ad expenditure to find the sales lift and essentially find ROI for advertising in Sunday circulars. Question: I am using weekly data, but for the ads can only get data back to 2001, so I have about 87 data points. Of those 87 data points, about 34 are weeks in which there was ad expenditure, while the remaining 53 weeks had no ads. When I get the coefficient of this ad expenditure variable, it will be an aggregate number, saying that over these 87 weeks, a dollar increase in ads results in an x% increase in unit sales. However, won't this percentage be understated since it is an aggregate number for the total 87 weeks? Can anyone help me think of another way to break it out? Does my questions make any sense? Last but not least, in interpreting the coefficients in the equation (below), how do you know when it is a percent? For example, I was assuming that a one-unit increase in x4 leads to a 1.1% increase in sales, but then what about x3? Does a one unit increase lead to a 524% decrease in sales or is it a 5 unit decrease?Thanks!
The `regression equation is weekly sales = - 704 + 0.543 x1 + 0.00597 x2- 5.24 x3 + 0.0111 x4 R-Sq(adj) = 90.0% . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
