Matching

[Md Sumon Ali]

Hello,

Hope this mail find you well. I am trying to use matching approach in my study. I am using the following stata code:

1. kmatch ps treated x1 x2 x3 x4, ate ematch(state county) gen

ii. kmatch md treated x1 x2 x3 x4, ate ematch(state county) gen

I successfully run these codes. But, I am encountering an issue when I run the regression after deleting the unmatched thing. In my regression, I am trying to use pweight=match weight, the model produces everything empty except coefficient. However, I have noticed that average match weight value is significantly high when treatment=1. For example, average value for control group is less than 1 but for treatment is 500+. It is worth mentioning that number of treatment is 45000 and control is 8 million

My question is should I avoid to use weight in regression? Or do you have any suggestion to me.

Thank you very much for your time. I look forward to your guidance.

[Felix Bittmann]

I am not sure what you mean with "I am encountering an issue when I run the regression after deleting the unmatched thing". Apparently you do not show all code, so please add it, otherwise it will be difficult to see what is going on. You do not need to run a separate regression afterwards as kmatch can compute the ATE for you as follows:

kmatch ps treated x1 x2 x3 x4 (outcome), ate ematch(state county)

However, if you want to run the regression on your own, you need to request the final weight as such:

kmatch ps treated x1 x2 x3 x4, ate ematch(state county) wgen(matchweight)
reg ...  [pweight=matchweight]

[Md Sumon Ali]

Thank you for your response. I have another question, which option should I choose ATT or ATE?

[Felix Bittmann]

I think the ATE is more general but you can easily report both.