When an independent variable is nonmetric (e.g., seasonality), and has k (two
or more) categories, it can be represented by k-1 dummy variables.
For example, the coding of in-store causal influences like display and co-op
advertising can be captured with the use 3 dummy variables:
- Display only. (No co-op): D1=1, D2=D3=0.
- Co-op, no display: D2=1, D1=D3=0.
- Display + co-op: D3=1, D1=D2=0.
- The 4th option, no display and no co-op, is represented by
D1=D2=D3=0.
The impact of dummy variables is reflected in the change of intercept (b0).
For instance, consider the relationship between sales and discount price. The x-coefficient
(b1 = discount price elasticity) remains unchanged when there is a display. The impact of
an in-store display is reflected in an incremental lift in sales, referred to as uplift.
