My Marketing Mix Modeling data contains some categorical variables which I convert into dummy variables. Before trying to create a Log-Log model I plan on replacing the 0 values with 1 and 1 values with 2.71828. My hypothesis is that this would make the dummy variable act as a filter which is on when the value is 2.71828 (1) and off when the value is 1 (0). Please let me know if this line of thinking is correct or not.
That will just making interpretting whatever coefficient trickier (if dividing by 2.72 is tricky). It is still a binary variable.
That will just making interpretting whatever coefficient trickier (if dividing by 2.72 is tricky). It is still a binary variable.
I think this will work, but you can just calculate logs first and then use the standard linear regression with dummies!
If the point is to remove a point singularity then there are a number of ways to get it done. One way, for example, would be to first use a simple linear transformation in the form of u=ax+1, and then to apply y=log(u). The final transform will depend on the exact problem that you are attempting to solve.
Is there a reason why you're taking the log of every one of your independent variables? Why can you not leave the binary variables as {0, 1} and just take the log of the continuous variables?
and to say is it right or wrong, some is really biased, those just mostly assumptions to say least, now if you have taken those substitutes it is, but how % of missing data, and central tendency of data set is? and if your variance is not much when you do it (by chaining) need to look from Mean and median point which applies on whole set .
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics