Is there anyway to conduct propensity matching score (by using psmatch2 or teffects psmatch) for panel data?

[Phuc Nguyen]

I am trying to do PSM then DiD for panel data. However, when searching online, I mostly see the example for cross-sectional data using psmatch2 or teffects psmatch.
I used to proposed to my professors that we can use two commonly-used approaches to convert panel data to cross-sectional data to deal with that problem (1) choosing the data of the year closest to the event date (2) getting average data of all years before event date.

However, my prof did not agree with that, they said that it may be suitable for a subsample test or extension, not for the main methodology because you cannot capture the difference in year, loss in information, standard error ,so on and so forth ...

I am wondering if there is any guidance or website that I can follow to run PSM for panel data.

Best regards.

[Jared Greathouse]

Before we talk methods, let's talk structure.


What intervention are you investigating? How long of a time series do you have? How many treated units and untreated units are there?

[Phuc Nguyen]

Before we talk methods, let's talk structure.


What intervention are you investigating? How long of a time series do you have? How many treated units and untreated units are there?

Thank Jared Greathouse , I am examining the tariff lifting up between two countries for custom barrier reduction and better supply chain. I have 6 years of data, 3 before and 3 after event date. There are around 600 treatment units and 5000 control units.

[Jared Greathouse]

Then your design seems fine, but my advice to you (knowing nothing about the further details) is to go back to the drawing board. You have a very very very short pre-intervention period, and I'm quite skeptical of any results even a well done PSM-DD would bring.

You say you're interested in tariffs on supply chains.... but what's the outcome exactly, how are we measuring supply chains? i think you need a longer pre-intervention series. I'm not saying it's at present impossible, but it's not desirable

[Phuc Nguyen]

Then your design seems fine, but my advice to you (knowing nothing about the further details) is to go back to the drawing board. You have a very very very short pre-intervention period, and I'm quite skeptical of any results even a well done PSM-DD would bring.

You say you're interested in tariffs on supply chains.... but what's the outcome exactly, how are we measuring supply chains? i think you need a longer pre-intervention series. I'm not saying it's at present impossible, but it's not desirable

Thank you so much, to me it is mostly a chance to combine DiD and PSM. Just run with all I have at hand

[Tom Scott]

Phuc Nguyen see https://arxiv.org/pdf/1812.01723.pdf. You might also consider an augmented synthetic control model - https://par.nsf.gov/servlets/purl/10331930