I am trying to use the propensity matching score to match the control and treatment groups based on their characteristics (can be done by gmatch, psmatch2, tseffect match...)
However, after matching, and even testing balance test, common support test, score plotting, there is another question coming to my mind: What should we do next to see how law implementation (event date) has impacts on treatment group
When reading a paper of Fan,2021 , I saw they use the term "DiD" after using PSM, which is normally known as treat*post variable in a standard two-way fixed effect Difference-in-Differences estimator. But from my point of view, whether we need that? I deem we only need the treat variable only instead of full combo of three conventional variables of difference-in-differences (post, treat, post*treat)
Sorry if it is not clear, I mean, for a panel data, normally how we run a two-way fixed effect DiD is
y= post + treat+ post*treat+ x1 + x2
And we explain the effect caused by law by the coefficient of post*treat variable
But from my point of view, after doing propensity matching score (especially 1:1), our variable of interest is treat, not treat*post. I mean, it is fair that we still put all term like that into regression but the difference caused by law now should be the coefficient of post rather than post*treat.
Please let me know if my explanation is unclear.
However, after matching, and even testing balance test, common support test, score plotting, there is another question coming to my mind: What should we do next to see how law implementation (event date) has impacts on treatment group
When reading a paper of Fan,2021 , I saw they use the term "DiD" after using PSM, which is normally known as treat*post variable in a standard two-way fixed effect Difference-in-Differences estimator. But from my point of view, whether we need that? I deem we only need the treat variable only instead of full combo of three conventional variables of difference-in-differences (post, treat, post*treat)
Sorry if it is not clear, I mean, for a panel data, normally how we run a two-way fixed effect DiD is
y= post + treat+ post*treat+ x1 + x2
And we explain the effect caused by law by the coefficient of post*treat variable
But from my point of view, after doing propensity matching score (especially 1:1), our variable of interest is treat, not treat*post. I mean, it is fair that we still put all term like that into regression but the difference caused by law now should be the coefficient of post rather than post*treat.
Please let me know if my explanation is unclear.
