# Difference-in-difference effect with triple interaction

I added an image of the regression output I could use some help with, as I am trying to recreate this but with data from a different time period.

This is the result of a difference-in-difference regression. Event, Affected and BHC are all dummy variables. The top part is just the standard output of the regression, the bottom part is added by the author, and this is what I'm interested in.

How is Event x Affected for BHC = 1 calculated? I'm pretty sure it is just the difference between Event x Affected and the triple interaction. But if it is just calculated by doing 0.0104-0.0089 why is it not 0.0015 and how would you go about getting the standard deviation (0.0019)?

• You should consider rounding for the first question, and the formula for the variance of the sum of random variables for the second. Also, note that this are standard errors, not standard deviations. – Dimitriy V. Masterov Sep 21 '18 at 4:25
• To answer this question with certainty, it would be great to explain in a few words the context (what is the outcome, how are affected, event and BHC defined) and to report all the main coefficients of the regression (I can imagine that there are other regressors in this regression?). – Roland Sep 21 '18 at 17:11

As you have guessed, the point estimate of the diff-in-diff effect is the sum between the coefficient on Event $$\times$$ Affected and the coefficient on Event $$\times$$ Affected $$\times$$ BHC. As these variables are not independent and the covariance of coefficients are not reported, it is not possible to infer the standard error of the diff-in-diff effect from the table only.
What a practitioner does to get directly the point estimates and standard errors of the quantities of interest is to change the specification of the regression. In this case, this means replacing "Event $$\times$$ Affected" by "Event $$\times$$ Affected $$\times$$ (BHC==0)" in the regressor list. If you do this, you'll get the diff-in-diff effects for independent banks (coefficient in front of "Event $$\times$$ Affected $$\times$$ (BHC==0)") and for BHC banks (coefficient in front of "Event $$\times$$ Affected $$\times$$ BHC").