Le mercredi 07 janvier 2015 à 14:19 +0100, Tamas Papp a écrit : > On Wed, Jan 07 2015, Milan Bouchet-Valat <[email protected]> wrote: > > > Le mercredi 07 janvier 2015 à 04:25 -0800, [email protected] a écrit : > >> A Nullable is immutable, its value isn't down the back of the couch > >> (which is my understanding of epistemological missingness, usually > >> applied to the TV remote :), it can never get a value once its null. > > That's not a technical question (immutable/mutable), but a conceptual > > one. If you have missing values in e.g. survey data, it usually means > > that the individual has not replied to the question (away, refused to > > reply, bug in the collect...). So you cannot say whether the value would > > have been 3 or something else. > > IMO it is very difficult to come up with a set of rules for operations > on missing data that satisfies all users (and uses), mostly because > "epistemological" and "ontological" missingness is sometimes mixed in > the same program/library, occasionally in subtle ways. > > When a first best solution is not possible, my preference is for > simplicity, which in this case means having a simple mental model of how > missingness works. If I understand Nullable correctly, there is one > simple rule to grok: "missingness propagates" -- that's it. I find this > appealing, even if I have to work around some corner cases. > > My understanding is that R is based on the same principle with respect > to NA, and it seems to work out (and, at the same time, is occasionally > confusing to newbies, but that may be inevitable). I'd be fine with that, but then other people seem to have a different idea of how Nullable should behave.
Regards
