I am running a regression that requires a fixed effect variable, which we’ll call fe. When I manually remove the singleton observations, I am left with 22,000 observations.

Here is my code for filtering singletons:
gen fe_filter = 0
by fe, sort: egen tot = count(fe)
replace fe_filter = 1 if tot > 1
keep if fe_filter == 1

I then use areg:

areg y x1, absorb(fe)

and get an r squared of .88.


I then run reghdfe, again after manually dropping the singletons, again leaving n=22000. However it automatically drops another 12000 observations, leaving 10,000 observations and an r squared of .75.

Here is my code for reghfe
reghdfe y x1, absorb(fe)

Any idea why regdhfe might be dropping the extra 12000 observations?