aweight, pweight or fweight? What to use for WLS?

[Simon Lesmeister]

Dear all,

I would like to apply a weighted least squares model but I do not know which weight to use for my problem. I am following the methodology of a published paper. The authors state: "We use weighted least squares, weighting each firm by the inverse of the number of firms in that country-quarter. This is to avoid giving excessive weight to countries in our sample that have a greater fraction of the number of firms such as the U.S.".


Now, I use the following code to get the number of distinct firms and after that to do my regression analysis:

egen tag = tag(firmid domicile_id qdate)
egen distinct = total(tag), by(domicile_id qdate)
gen weight = 1/distinct


reghdfe y x [pweight=weight], vce(cluster firmid)

Is it pweight, aweight or fweight to use in this context? Is it correct to use my variable "weight", or do I have to use the variable "distinct" in the square brackets?


Thank you very much in advance and best wishes,
Simon

[FernandoRios]

Hi Simon
Definitely, fweight will not work here, as it only admits weights without decimals.
aweights is the one that will provide you with the standard WLS (as what you would do in a standard textbook).
However, I would also consider using pweights, to get Robust standard errors
In any case, if you use cluster option, it does not matter if you use aweights or pweghts.
Best regards