[Gretl-users] Re: $sample accessor in a WLS context
On Sat, 16 Jan 2021, Artur Bala wrote:
Dear,
In a WLS model observations with weight=0 are excluded from the estimation
and gretl acknowledges this by reporting on "missing or incomplete
observations dropped".
However, the "$sample" accessor doesnt' account for these
"missing"
observations
and still assigns them a 1 value. Technically these observations are not
part of the estimation sample and logically they should correspond to a 0
value in the "$sample", shouldn't they?
Well, that's arguable. By the same logic, you should exclude from the
sample an observation whose weight is 1.0e-12. The way I see the $sample
accessor is "what this observation explicitly excluded as a result of a
sample restriction or missing observations?"; if this is not the case,
then the influence of that particular observation on the result could well
be 0, on account of the mechanics of the estimator you're computing, but
the observation is in the sample alright.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------