[Gretl-users] Re: 2021b git: WLS estimation bug

Am 17.04.2021 um 09:13 schrieb Artur Bala: > Dear Allin, > > In a WLS estimation when the weight variable contains zeros gretl stops > executing saying that "Weight variable is not a dummy but contains zeros" > The same worked perfectly before with the observations weighted 0 being > simply dropped > from the estimation. > Yes, that is an intentional change, because zero weights mess up the definition/concept of the sample size and so on. If you want to exclude certain observations, you now will have to explicitly restrict the sample to achieve the same effect. Or, if your choice of zero was just a shortcut to mean "really small", you can assign a "really small" but strictly positive weight instead. To put it differently, we reckoned that the earlier behavior could have been considered a (minor) bug.