Difference in differences

[Taiba Chau]

b

[Jared Greathouse]

Sounds good.

[Jared Greathouse]

EDIT: the DDD approach in general sounds like a plausible one. Without code or a better asked/formulated question, I otherwise don't endorse anything else about this.

[Taiba Chau]

Hi, this is my regression and output:

I was just wondering then the coefficient estimate for the interaction term for the country, wave and ethnic individual in each of the 9 countries. Essentially I am trying to see the different impact of the immigration policy on ethnic and non-ethnic individuals in the selected countries.

[Jared Greathouse]

I will be honest: DDD is basically a three way interaction term between policy, time and (usually) subgroup, and I'm no good with 3 way interactions, and neither is my methods teacher.

I've only done the standard DD setup, and I'm not qualified to talk about a triple differences design in the same way that I could a normal DD approach.

[Taiba Chau]

That is fine, i still appreciate the help. What if I wanted to do an ordinary DD model. How would that differ? Or how would that be implemented in this context? Thanks

[Clyde Schechter]

If you want the difference in policy effect between ethnic and non-ethnic in, say, Denmark, that is the sum of the following regression coefficients: 2.wave#1.ethnic + Denmark#2.wave#1.ethnic. To get that calculated along with its standard error, confidence interval, and test statistics, you can use the -lincom- command. The exact code for that depends on the names that Stata uses to refer to those coefficients in your _b[] matrix after the regression. And that, in turn, can vary among versions of Stata. So, to get the -lincom- syntax right, you have to re-run the regression adding the -coefl- option at the end of the command. Stata will re-run the regression, this time showing the coefficients and their names inside the _b[] matrix. Use that information to write your -lincom- command.

[Taiba Chau]

That was very helpful thank you!