CODEWITHSUNDEEP
stata
20170524
20170524

Reputation: 27

Stata difference in difference univariate tests

I want to calculate univariate t-statistic for difference-in-difference.

I have unbalenced panel dataset including treatment and control groups, as follows:

                  Treatment   Control  

 Pre-period         5.33       8.9     

 Post-period        5.10       6.9

From the above database, I would like to get

                  Treatment   Control  Difference

 Pre-period         5.33       8.9     -3.57

 Post-period        5.10       6.9     -1.8

 Difference         0.23       2.0     -1.77

Of course, I would like to get t-statistics for each of differences (in total, 5).

Thank you in advance.

Upvotes: 0

Views: 799

Answers (1)

chan1142
chan1142

Reputation: 643

Let

    D(,pre) = [Treat,Pre]    - [Control,Pre]
   D(,post) = [Treat,Post]   - [Control,Post]
  D(treat,) = [Treat,Post]   - [Treat,Pre]
D(control,) = [Control,Post] - [Control,Pre]
        DID = D(post) - D(pre) = D(treat) - D(control).

In your example, D(,pre) = -3.57, D(,post) = -1.8, D(treat,) = -0.23, D(control,) = -2.0 and DID = 1.77. (Note: We go post-pre rather than pre-post.) Let "Treat", "Control", "Pre" and "Post" be the four dummy variables.

  1. reg y Treat##Post gives D(,pre), D(control,) and DID.
  2. reg y Control##Post gives -D(,pre), D(treat,) and DID.
  3. reg y Treat##Pre gives D(,post), -D(control,) and DID.
  4. reg y Control##Pre gives -D(,post), -D(treat,) and DID.

Explanation: Do reg y Treat Post c.Treat#c.Post. The coefficient of Treat is E(y|Treat=1, Post=0) - E(y|Treat=0, Post=0), which is D(,pre). The coefficient of Post is E(y|Treat=0, Post=1) - E(y|Treat=0, Post=0), which is D(control,). The coefficient on the interaction term is DID obviously. Others are similar.

Upvotes: 1

Related Questions