PSM-FE with 2 year panel data

[Kriti Sharma]

Dear all,

I am trying to run fixed effects regression after conducting PSM on a panel data of two years. However, fixed effects requires weights to be identical within IDs. How should I modify it or proceed with it? My data has 489 treated values out of 2705 total observations in the first year (2018), and 579 treated values out of 2705 in the second year (2023). Those treated (having access to a government scheme) in the first year may or may not be treated in the second year and vice versa. I ran PSM on both years, should I use for one year? Is it possible to conduct the analysis with the same data structure?

[George Ford]

So you've got treated in period 1 that might stay treated or switch to untreated in period 2. And, you've got untreated in period 1 that may stay untreated or switch to treated in period 2. The switchers identify the effects.

With DID, you'd match on the pre-treatment values. But this is not a straightforward DID.

Maybe create a "is ever treated" variable and match on that in year 1, then copy the weights over to year 2. (and maybe the other way around to see if it changes the results by much).

This is interesting and I hope some others offer suggestions.

Show us some results.

I suppose in the cleanest case possible the switching coefficients should be equal in absolute value, but I doubt you'd see that since its a "government scheme" and the effect may linger.

Might be interesting to estimate year-by-year for the cross-sectional effect and then compare that to the size of the effects for the switchers.

Or, drop all treated in period 1 and do a 2x2 DID on the treated in T2.

And, included a "to be treated" dummy in period 1 and estimate for period 1 as a pseudo treatment (selection bias). I'd think if the coef on the future treatment is zero, then you may not need to match.

Lot's of fun here.

[Kriti Sharma]

Dear George,

Thank you for your response. I will try to think on these lines. Hoping some others can help out on the forum.

Regards,
Kriti