Cluster standard errors

[Nirosha Dissanayaka]

Hi all,
I am using district level data with difference in differences method.
(One treated district and 24 untreated districts)
Data comes from sample survey. But not panel data.
Data for the period of 2005-2017

I use district and year fixed effects.

Can anyone let me know should I use robust standard errors or cluster standard errors?

Thank you in advance..

[Fei Wang]

Nirosha, I'm a bit confused. I think you are using a district-level panel data with 25 districts and 13 years (325 observations). Why did you say it's not a panel data? Or did you mean it's an individual-level data from 25 districts and 13 years, but it's a pooled cross section at individual level?

[Maxence Morlet]

With one treated district and 24 untreated, I would rather use synthetic control method if I were you (e.g. Cerulli's 2020 package npsynth). Also note that the vce(cluster unitID) option will produce heteroscedasticity-robust standard errors clustered at the unit level.