Thank you for reading this distress post.
I have a panel dataset with store locations as group variable, and the date as a time variable.
It has 3 explanatory variables:
i.dow (day of week)
discountday (A boolean variable that shows if that specific store runs a discount for the product on that specific day. note that a discount can only be on a Friday)
i.anticipation ( variable that shows how many days before/after the discount day is, only for the store that has the discount. It can take the values -4 until 4, where the Thursday prior to the sale is -1, the Wednesday is -2 etc. remaining observation values are blank.) (this is the objective of my study, so i have to include it)
This obviously causes a lot of collinearity, since for example discountday can only occur on i.dow ==6, and i.anticipation ==-1 can only coincide with i.dow==4.
However, I do need all these regressors to have a good model.
I made a control group by dropping all values of discountday==1, and all values of i.anticipation that are not blank.
Then used:
Code:
xtreg y i.dow, fe
Is it possible to 'lock' these values, add back the dropped observations,
and then use the other two independent variables AFTER this first command, while retaining the i.dow values from xtreg command?
I have no idea how to do something like this, every help is much appreciated.
Also, if there is a better way to achieve this, I would be interested in hearing that too.
Kind regards, a long term lurker but now new member
Theo
0 Response to Work multicollinearity out of fe model.
Post a Comment