Currently, the documentation is silent on standard error defaults - would be good to add notes to the feols() and fepois() docs.
In alignment with fixest, the behavior is the following:
If no fixed effect is specified to be projected out, feols() and fepois() return "iid" standard errors.
If one (or more) fixed effects are specified via formula syntax, inference is clustered by the first fixed effect (CRV1 inference).
feols("Y ~ X + C(f1)", data = data) # no fixed effect specified via two-part formula syntax -> iid inference
feols("Y ~ X | f1", data = data) # fixed effect specified via two-part formula syntax -> CRV1 inference clustered by f1
feols("Y ~ X | f1 + f2", data = data) # two fixed effects specified via two-part formula syntax -> CRV1 inference clustered by the first fixed effect f1
To do
Add this information to the standard error section in the quickstart notebook.
Add this information to the docstrings of the vcov of feols() and fepois().
Context
Currently, the documentation is silent on standard error defaults - would be good to add notes to the
feols()
andfepois()
docs.In alignment with
fixest
, the behavior is the following:feols()
andfepois()
return "iid" standard errors.To do
vcov
offeols()
andfepois()
.