DID analysis including time-varying covariates

[Kyle Lee]

Hi. I'm studying Difference-in-Differences and have read Zeldow&Hatfield(2021) that is about the method of correcting bias due to observed confounders.
In the article, I understand that when there are time-varying covariates which means evolve differently by group, they confound the treatment effect so that you must adjust for the interaction of a covariate and time(time-varying adjusted model).
Time-varying adjusted model allows the coefficient on the covarite to vary over time.

I know the commands 'didregress' and 'xtdidregress' fit DID models in STATA17.
But I'm not familiar with the commands yet.

Do the commands 'didregress' and 'xtdidregress' can fit time-varying adjusted DID models?


(I'm not good at english. I'm sorry if you cannot understand my writing and if needed, I would explain or add some details.)

[Jared Greathouse]

I understand you fine, but I don't understand the problem. You can add covariates to xtdidreg if you choose to.

[Kyle Lee]

Thank you for answering, Mr. Greathouse.

In the article(Zeldow&Hatfield, 2021),
the covariate adjusted model is adjusted for the covariate with a constant coefficient(λx_it)
and the time-varying adjusted model allows the coefficient on the covariate(λ_tx_it) to vary over time.

I want to fit the time-varying adjusted model with STATA.
But the model to fit with the command '(xt)didregress' is like below according to "STATA TREATMENT-EFFECTS REFERENCE MANUAL" distributed by STATA Corp.,

y_ist = γ_s + γ_t + z_istβ + D_stδ + ε_ist

In the model, the coefficient of covariates is not "β_t" but "β". So, I understood the model is not adjusted for time-varying confounders.

[Jared Greathouse]

Just use an interaction term between time and your predictor

i.time##c.var

[Kyle Lee]

Thank you very much, Mr. Greathouse!
Your reply is very helpful.