Problems when implementing PSM (propensity score match) and DID (difference in difference)

[Ni Li]

Hi,
I am using Stata 12 and I have a panel data. Sample period of this data is from 2004 to 2013.

I have yearly observations for a treated and control group. I want to generate a matched sample so to compare the difference of outcome between observations from treatment group and control group before and after the treatment events 2 years respectively. The year of treatment event is during 2006 to 2011. Hence, I need to use the propensity score match combined with difference in difference test. However, I faced some problems when implementing these procedures in Stata, including:

1. Observations in treatment group are treated in different years, for example, firm A was treated in 2010 and firm B was treated in 2008. In this case, it is possible that the treatment sample in this year will become the control sample in another year. What can I do to cope with this problem?
2. Observations in treatment group are treated in different years. However, if I use psmatch2 to match the sample, I need to define the dummy variable of period in unified way and I am confused on this.
3. Stata provide the command of diff for combining PSM and DID in one step. However, this syntax can only match the sample with kernel match. If I want to match the sample in other ways, such as nearest neighbor match, I need to calculate the weight for matching and edit the program by myself. However, I don't know how to execute these procedures now and still look for the solutions.

Thanks everyone. Looking forward to your kindly suggestions and your help will be appreciated

[natiesim]

Hello, Please have you got a solution to your problem?

[Ma Zhong]

I face similar situation you mentioned in (2). Recently I read a working paper “Mandatory CSR Disclosure and Shareholder Value: Evidence from China”. I feel puzzled in their PSM approach.
In their Appendix B “Procedure to develop propensity-score-matched sample”, it seems that they only do one Logit regression with pre-period samples. However, how do they pick the control samples matching in post period?
I still get no reply for this question now.

[bishal bharadwaj]

Dear Ni Li,

I am facing similar problem in developing difference in difference data using PSM from set of before-after and another dataset. however, i don't know how to execute it?
Thank you for your suggestion