# Running a regression to avoid multicollinearity

I have the following regression (pooled OLS; panel data):

Y  Treated Shock Shock*Treated {with industry and year fixed effects}


Y is a continuous variable, “Treated” is a count variable between 1 and 10, which does not change within the years. “Shock” is equal to 1 if year>1995 and zero otherwise. “ShockTreated” is multiplication of ShockTreated. Below I provide you with the example of how data looks like for 1 firm:

Once I run the regression above the results omit “Shock” variable. I guess the problem is that I include year and industry fixed effects. Shall I control just for the industry fixed effect? I simply dont know which fixed effects (dummies for years and industry) to include so that all the variables remain (i.e, not dropped after I run the regression

• It is likely no one will be able to tell you what to do without knowing what your goal is. However unless treated is always 10 when Shock is 1, this does not seem to be multicollinearity. – Giskard Oct 15 at 18:52
• What exactly do you mean by year fixed effects? Dummy variables for each year? – Giskard Oct 15 at 18:54
• Thank you for your reply. I simply dont know which fixed effects to include so that all the variables remain (i.e, not dropped after I run the regression – Alberto Alvarez Oct 15 at 18:54
• Fixed effects - dummies for each year and for each industry – Alberto Alvarez Oct 15 at 18:54
• Shock is quite clearly 1 - the sum of the before 1995 year dummies, so you cannot include those and schock as well. – Giskard Oct 15 at 18:56