Difference-in-differences results interpretation

[Sacha Elinemez]

Dear all,

I have run a difference-in-differences regression and now I am struggling to interpret the results.
I have used the following variables:

Treatment = (price / total revenue)
Post = 1 for years post-event, 0 for years pre-event.
Interaction variable = Treatment * Post
Dependent variable = (Earnings before interest and taxes / lagged total assets)

Now the coefficient that I get for the interaction variable regressed on the dependent variable is 0.110.
How do I interpret this?

Thank you for your help.

[Oscar Ozfidan]

in the post period when post dummy is 1, the treatment dependent variable relationship is higher by 0.11 than the same relationship in the pre-event sample. so pre-event relationship is the coefficient of Treatment. Post event relationship is the coefficient of Treatment (pre-event) plus the coefficient of interaction term

[Sacha Elinemez]

Hi Oscar Ozfidan,

Thank you for your reply.

Does this mean that the (EBIT/ lagged total assets) increases with 11.0% after the event when the price relative to total revenue increases with 1%?

[Jeff Wooldridge]

Sacha: Is that a continuous treatment variable? I'ts price/total revenue, right? Not a binary variable. What would it mean for price/total to increase by one unit? Assuming price and earnings are always strictly positive, I would take the logs. Or, I would take the log of (price/revenue) and use Poisson regression for earnings and control for lagged total assets -- actually log(total assets).

But, as in the FAQ, you'll get a better answer if you post your Stata commands and Stata output.

[Oscar Ozfidan]

Hi Oscar Ozfidan,

Thank you for your reply.

Does this mean that the (EBIT/ lagged total assets) increases with 11.0% after the event when the price relative to total revenue increases with 1%?

Like Jeff Wooldridge said you will get better traction in this forum if you post a sample of data. The guiding principle is "for every one unit change in (price / total revenue), (Earnings before interest and taxes / lagged total assets) appears to change by 0.11 unit." To me that sound like approx one tenth of the change in independent variable. But exact interpretation depends entirely whether those variables are percentages or not. Based on the information you provided they appear to be. Also, it would be extremely unlikely to see a ten times more change in dep var when you change the indep var by one unit. Those are my educated guesses but I know I could be wrong too. Just stick to the principle I stated and you will reach to the right conclusion.

[Sacha Elinemez]

Thank you very much for the replies!