Difference-in-Difference on panel data without treatment and control group distinction

[Efe Daniel]

I am writing a thesis (MSc level) on costs of a particular regulation on business in one country (pharmaceutical company). My independent variable is compliance costs (yearly). Explanatory variables are dummy variable of 1 for after regulation and zero for before regulation and various accounting data such as total assets, EBIT, cash flow measures, debt to equity at period end to control for firm size. All observations are yearly. I need to estimate the costs of the regulation using difference in difference (see page 12 of “Internet Appendix to “The Effect of SOX Section 404: Costs, Earnings Quality and Stock Prices””, download from the internet. I do not understand how the author applied the method. I urgently need assistance on how i can apply the DiD method on my research.

[Jorge Eduardo Perez Perez]

I do not understand how the author applied the method. I urgently need assistance on how i can apply the DiD method on my research.

Maybe if you provide more details we can help, but this is not enough. Let us know what dataset you're trying to use, if it is panel data, details about the policy and what you've tried in Stata so far. Keep in mind that most people are not willing to read through this paper or even search for it.

[Enrique Pinzon (StataCorp)]

Dear Efe,

Jeff Wooldridge in his Econometrics of Cross Section and Panel Data has a nice discussion about the topic in Chapter 6 section 5 (6.5). I will describe the estimation procedure he suggests below. Define the following variables:

dm - Dummy variable that equals 1 if there was a change in regulation and 0 otherwise.
dt - Dummy variable that equals 1 for the time period after the policy and 0 for the time period before the policy.
x - Other control variables

The most basic model is given by:
\begin{equation} y = \beta_0 + \beta_1dm + \beta_2dt + \beta_3 dt \times dm + \beta_4 x + \varepsilon \end{equation} The coefficient for the interaction (dm x dt) is the one that captures the effect of the change in regulation. In Stata I would type:

 regress y i.dt##i.dm x, vce(cluster id)

Note that I cluster my standard errors at the panel data level (id) when I type vce(cluster id).