6:12 p.m.
Hi,
I have used the "^" trick as described in the guide to create
interaction effects quickly. In particular my interaction variable is a
month indicator in a monthly time series workfile.
However, the resulting interaction series have valid entries where the
original series had missings (beyond the latest available obs), but only
where the interaction effect would be "active" (=non-zero).
For example, the interaction of x with January would have a missing at
obs 2016:1 but would be zero for 2016:2, although x has a missing at
both of these obs.
I think this is a bug. I can understand the logic of "no matter what the
value, the interaction is going to be zero if the month != 1", but I
think missings should always propagate. Or do we have precedents of the
opposite in gretl?
thanks,
sven
8:53 p.m.
On Tue, 19 Jul 2016, Sven Schreiber wrote:
Hi,
I have used the "^" trick as described in the guide to create interaction
effects quickly. In particular my interaction variable is a month indicator
in a monthly time series workfile.
However, the resulting interaction series have valid entries where the
original series had missings (beyond the latest available obs), but only
where the interaction effect would be "active" (=non-zero).
For example, the interaction of x with January would have a missing at obs
2016:1 but would be zero for 2016:2, although x has a missing at both of
these obs.
I think this is a bug. I can understand the logic of "no matter what the
value, the interaction is going to be zero if the month != 1", but I think
missings should always propagate. Or do we have precedents of the opposite in
gretl?
In gretl, we stipulate that zero times NA is zero (which, I have
argued, makes sense so laong as "NA" means unobserved, not an
arithmetical pathology such as NaN or infinity). But if this case is
going beyond that stipulation then I suppose it is a bug. If you
could come up with a minimal test case that would be very helpful.
Allin
2:30 p.m.
Am 19.07.2016 um 22:53 schrieb Allin Cottrell:
On Tue, 19 Jul 2016, Sven Schreiber wrote:
>
> However, the resulting interaction series have valid entries where the
> original series had missings (beyond the latest available obs), but
> only where the interaction effect would be "active" (=non-zero).
(Sorry, I meant it the other way around...)
> For example, the interaction of x with January would have a
missing at
> obs 2016:1 but would be zero for 2016:2, although x has a missing at
> both of these obs.
(... namely like in this example.)
In gretl, we stipulate that zero times NA is zero (which, I have
argued,
makes sense so laong as "NA" means unobserved, not an arithmetical
pathology such as NaN or infinity). But if this case is going beyond
that stipulation then I suppose it is a bug.
Right, now I remember some discussions about this. This is actually a
case that follows that stipulation.
thanks,
sven
2:05 p.m.
New subject: gretl crash
Open the attached data, and try to do a Butterworth filter via GUI.
Gretl crashes.
If I do the same via console command it works OK.
I am running gretl from current git on Ubuntu Linux 14.04
--
Ignacio Díaz-Emparanza
Departamento de Economía Aplicada III (Econometría y Estadística)
Universidad del País Vasco - Euskalherriko Unibertsitatea, UPV/EHU
Tfno: (+34) 94 601 3732
http://www.ehu.eus/ea3
2:59 p.m.
New subject: gretl crash
On Thu, 21 Jul 2016, Ignacio Diaz-Emparanza wrote:
Open the attached data, and try to do a Butterworth filter via GUI.
Gretl
crashes.
If I do the same via console command it works OK.
I am running gretl from current git on Ubuntu Linux 14.04
Thanks for the report, Ignacio. That's now fixed in git. The crash was
due to a very long series-name plus a translated string overflowing a
character buffer of fixed length (in composing the filter plot). The
relevant buffers now get sized dynamically.
Allin
3077
days inactive
3079
days old
4 comments
3 participants
participants (3)
-
Allin Cottrell -
Ignacio Diaz-Emparanza -
Sven Schreiber